@hasna/evals 0.1.20 → 0.1.22

package/LICENSE

@@ -1,3 +1,4 @@
+
                                  Apache License
                            Version 2.0, January 2004
                         http://www.apache.org/licenses/
@@ -175,7 +176,7 @@
 
    END OF TERMS AND CONDITIONS
 
-   Copyright 2026 hasna
+   Copyright 2026 Hasna, Inc.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.

package/README.md

@@ -240,6 +240,19 @@ evals_run_single(
 → PASS — The response correctly identifies Paris.
 ```
 
+## HTTP mode
+
+Shared Streamable HTTP transport for multi-agent sessions (stdio remains the default):
+
+```bash
+evals-mcp --http              # http://127.0.0.1:8817/mcp
+MCP_HTTP=1 evals-mcp          # same
+evals-mcp --http --port 8817  # explicit port
+```
+
+- Health: `GET http://127.0.0.1:8817/health` → `{"status":"ok","name":"evals"}`
+- Override port with `MCP_HTTP_PORT` or `--port`
+
 ---
 
 ## License

package/dist/cli/commands/compare.d.ts

@@ -1,3 +1,5 @@
 import { Command } from "commander";
+import { compareRuns } from "../../core/reporter.js";
+export declare function renderMarkdownDiff(diff: ReturnType<typeof compareRuns>): string;
 export declare function compareCommand(): Command;
 //# sourceMappingURL=compare.d.ts.map

package/dist/cli/commands/compare.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"compare.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/compare.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAIpC,wBAAgB,cAAc,IAAI,OAAO,CAgCxC"}
+{"version":3,"file":"compare.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/compare.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,WAAW,EAA+B,MAAM,wBAAwB,CAAC;AAGlF,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,UAAU,CAAC,OAAO,WAAW,CAAC,GAAG,MAAM,CA2B/E;AAED,wBAAgB,cAAc,IAAI,OAAO,CAgCxC"}

package/dist/cli/commands/compare.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=compare.test.d.ts.map

package/dist/cli/commands/compare.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"compare.test.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/compare.test.ts"],"names":[],"mappings":""}

package/dist/cli/commands/doctor.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"doctor.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/doctor.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAqCpC,wBAAgB,aAAa,IAAI,OAAO,CA4EvC"}
+{"version":3,"file":"doctor.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/doctor.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAqCpC,wBAAgB,aAAa,IAAI,OAAO,CAqFvC"}

package/dist/cli/commands/generate.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"generate.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/generate.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAiBpC,wBAAgB,eAAe,IAAI,OAAO,CA2CzC"}
+{"version":3,"file":"generate.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/generate.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAiBpC,wBAAgB,eAAe,IAAI,OAAO,CAyDzC"}

package/dist/cli/index.js

@@ -30508,7 +30508,7 @@ function parseAdapterConfig(opts) {
 
 // src/cli/commands/run.ts
 function runCommand() {
-  return new Command("run").description("Run an eval dataset against your app").argument("<dataset>", "Path to JSONL/JSON dataset or glob pattern").option("--adapter <type>", "Adapter type: http|anthropic|openai|mcp|function|cli", "http").option("--url <url>", "App URL (for http adapter)").option("--model <model>", "Model name (for anthropic/openai adapter)").option("--system <prompt>", "System prompt (for anthropic/openai adapter)").option("--module <path>", "Module path (for function adapter)").option("--export <name>", "Export name (for function adapter, default: default)").option("--command <cmd>", "Shell command (for cli adapter, use {{input}} placeholder)").option("--mcp-command <cmd>", "MCP server command (for mcp adapter)").option("--tool <name>", "MCP tool name (for mcp adapter)").option("--concurrency <n>", "Parallel execution limit", "5").option("--repeat <n>", "Run each case N times (Pass^k metric)", "1").option("--tags <tags>", "Comma-separated tags to filter cases").option("--no-judge", "Skip LLM judge, run assertions only").option("--output <format>", "Output format: terminal|json|markdown", "terminal").option("--save", "Save run to database").option("--json", "Alias for --output json").action(async (dataset, opts) => {
+  return new Command("run").description("Run an eval dataset against your app").argument("<dataset>", "Path to JSONL/JSON dataset or glob pattern").option("--adapter <type>", "Adapter type: http|anthropic|openai|mcp|function|cli", "http").option("--url <url>", "App URL (for http adapter)").option("--model <model>", "Model name (for anthropic/openai adapter)").option("--system <prompt>", "System prompt (for anthropic/openai adapter)").option("--module <path>", "Module path (for function adapter)").option("--export <name>", "Export name (for function adapter, default: default)").option("--command <cmd>", "Shell command (for cli adapter, use {{input}} placeholder)").option("--mcp-command <cmd>", "MCP server command (for mcp adapter)").option("--tool <name>", "MCP tool name (for mcp adapter)").option("--concurrency <n>", "Parallel execution limit", "5").option("--repeat <n>", "Run each case N times (Pass^k metric)", "1").option("--tags <tags>", "Comma-separated tags to filter cases").option("--no-judge", "Skip LLM judge, run assertions only").option("--output <format>", "Output format: terminal|json|markdown", "terminal").option("--save", "Save run to database").option("-j, --json", "Alias for --output json").action(async (dataset, opts) => {
     const { cases, warnings } = await loadDataset(dataset, {
       tags: opts["tags"] ? opts["tags"].split(",") : undefined
     });
@@ -30548,7 +30548,7 @@ init_loader();
 init_store();
 function ciCommand() {
   const cmd = new Command("ci").description("Run evals in CI mode \u2014 exit non-zero on regression");
-  cmd.command("run <dataset>").description("Run and compare to baseline").option("--adapter <type>", "Adapter type: http|anthropic|openai|mcp|function|cli", "http").option("--url <url>", "App URL (for http adapter)").option("--model <model>", "Model name (for anthropic/openai adapter)").option("--system <prompt>", "System prompt (for anthropic/openai adapter)").option("--module <path>", "Module path (for function adapter)").option("--export <name>", "Export name (for function adapter)").option("--command <cmd>", "Shell command (for cli adapter)").option("--mcp-command <cmd>", "MCP server command (for mcp adapter)").option("--tool <name>", "MCP tool name (for mcp adapter)").option("--no-judge", "Skip LLM judge, assertions only").option("--baseline <name>", "Baseline name to compare against", "main").option("--fail-if-regression <pct>", "Fail if score drops by more than N%", "0").option("--output <format>", "Output format: terminal|markdown", "terminal").option("--json", "Output JSON").action(async (dataset, opts) => {
+  cmd.command("run <dataset>").description("Run and compare to baseline").option("--adapter <type>", "Adapter type: http|anthropic|openai|mcp|function|cli", "http").option("--url <url>", "App URL (for http adapter)").option("--model <model>", "Model name (for anthropic/openai adapter)").option("--system <prompt>", "System prompt (for anthropic/openai adapter)").option("--module <path>", "Module path (for function adapter)").option("--export <name>", "Export name (for function adapter)").option("--command <cmd>", "Shell command (for cli adapter)").option("--mcp-command <cmd>", "MCP server command (for mcp adapter)").option("--tool <name>", "MCP tool name (for mcp adapter)").option("--no-judge", "Skip LLM judge, assertions only").option("--baseline <name>", "Baseline name to compare against", "main").option("--fail-if-regression <pct>", "Fail if score drops by more than N%", "0").option("--output <format>", "Output format: terminal|markdown", "terminal").option("-j, --json", "Output JSON").action(async (dataset, opts) => {
     const { cases } = await loadDataset(dataset);
     const adapter = parseAdapterConfig(opts);
     const run = await runEvals(cases, { dataset, adapter, skipJudge: opts["judge"] === false || opts["noJudge"] === true });
@@ -30596,7 +30596,7 @@ No baseline "${baselineName}" found \u2014 use "evals ci set-baseline" to create
 
 // src/cli/commands/judge.ts
 function judgeCommand() {
-  return new Command("judge").description("Ad-hoc: judge a single input/output pair against a rubric").requiredOption("--input <text>", "The input that was given to the AI").requiredOption("--output <text>", "The AI's response").requiredOption("--rubric <text>", "Plain-English grading criteria").option("--expected <text>", "Expected behavior description").option("--model <model>", "Judge model", "claude-sonnet-4-6").option("--provider <provider>", "Judge provider: anthropic|openai", "anthropic").option("--json", "Output JSON").action(async (opts) => {
+  return new Command("judge").description("Ad-hoc: judge a single input/output pair against a rubric").requiredOption("--input <text>", "The input that was given to the AI").requiredOption("--output <text>", "The AI's response").requiredOption("--rubric <text>", "Plain-English grading criteria").option("--expected <text>", "Expected behavior description").option("--model <model>", "Judge model", "claude-sonnet-4-6").option("--provider <provider>", "Judge provider: anthropic|openai", "anthropic").option("-j, --json", "Output JSON").action(async (opts) => {
     const result = await judgeOnce({
       input: opts["input"] ?? "",
       output: opts["output"] ?? "",
@@ -30622,8 +30622,34 @@ ${icon} ${result.verdict}
 
 // src/cli/commands/compare.ts
 init_store();
+function renderMarkdownDiff(diff) {
+  const lines = ["## Diff", ""];
+  if (diff.regressions.length === 0 && diff.improvements.length === 0) {
+    lines.push("- No changes between runs.");
+    return lines.join(`
+`);
+  }
+  if (diff.regressions.length > 0) {
+    lines.push("### Regressions");
+    for (const r of diff.regressions) {
+      lines.push(`- ${r.caseId}: ${r.before} -> ${r.after}`);
+    }
+    lines.push("");
+  }
+  if (diff.improvements.length > 0) {
+    lines.push("### Improvements");
+    for (const i of diff.improvements) {
+      lines.push(`- ${i.caseId}: ${i.before} -> ${i.after}`);
+    }
+    lines.push("");
+  }
+  const delta = diff.passRateDelta * 100;
+  lines.push(`- Score delta: ${delta >= 0 ? "+" : ""}${delta.toFixed(1)}%`);
+  return lines.join(`
+`);
+}
 function compareCommand() {
-  return new Command("compare").description("Compare two eval runs side-by-side").argument("<before>", "Before run ID or baseline name").argument("<after>", "After run ID (or 'latest')").option("--json", "Output JSON diff").option("--markdown", "Output markdown diff").action(async (beforeArg, afterArg, opts) => {
+  return new Command("compare").description("Compare two eval runs side-by-side").argument("<before>", "Before run ID or baseline name").argument("<after>", "After run ID (or 'latest')").option("-j, --json", "Output JSON diff").option("--markdown", "Output markdown diff").action(async (beforeArg, afterArg, opts) => {
     const { listRuns: listRuns2 } = await Promise.resolve().then(() => (init_store(), exports_store));
     const beforeRun = getRun(beforeArg) ?? getBaseline(beforeArg);
     const afterRun = afterArg === "latest" ? listRuns2(1)[0] : getRun(afterArg) ?? getBaseline(afterArg);
@@ -30640,10 +30666,8 @@ function compareCommand() {
       console.log(JSON.stringify(diff, null, 2));
     } else if (opts["markdown"]) {
       console.log(toMarkdown(afterRun));
-      console.log(`
-## Diff
-`);
-      printDiffReport(diff);
+      console.log();
+      console.log(renderMarkdownDiff(diff));
     } else {
       printDiffReport(diff);
     }
@@ -30662,7 +30686,7 @@ var COST_PER_1K_INPUT = {
 };
 var AVG_JUDGE_TOKENS = 800;
 function estimateCommand() {
-  return new Command("estimate").description("Estimate cost before running evals (no API calls made)").argument("<dataset>", "Path to JSONL/JSON dataset").option("--model <model>", "Judge model", "claude-sonnet-4-6").option("--no-judge", "Assume no judge calls").option("--json", "Output JSON").action(async (dataset, opts) => {
+  return new Command("estimate").description("Estimate cost before running evals (no API calls made)").argument("<dataset>", "Path to JSONL/JSON dataset").option("--model <model>", "Judge model", "claude-sonnet-4-6").option("--no-judge", "Assume no judge calls").option("-j, --json", "Output JSON").action(async (dataset, opts) => {
     const { cases, warnings } = await loadDataset(dataset);
     if (warnings.length > 0)
       for (const w of warnings)
@@ -30713,7 +30737,7 @@ Each case must be a valid JSON object on a single line with these fields:
 Generate varied cases that cover edge cases, typical usage, and boundary conditions.
 Output ONLY valid JSONL \u2014 one JSON object per line, no markdown, no explanation.`;
 function generateCommand() {
-  return new Command("generate").description("Generate eval cases from a description using Claude").requiredOption("--description <text>", "What behavior to test (e.g. 'refund policy responses')").option("--seeds <path>", "Path to JSONL file with seed examples").option("--count <n>", "Number of cases to generate", "10").option("--output <path>", "Output JSONL file path", "generated.jsonl").option("--model <model>", "Model to use for generation", "claude-sonnet-4-6").action(async (opts) => {
+  return new Command("generate").description("Generate eval cases from a description using Claude").requiredOption("--description <text>", "What behavior to test (e.g. 'refund policy responses')").option("--seeds <path>", "Path to JSONL file with seed examples").option("--count <n>", "Number of cases to generate", "10").option("--output <path>", "Output JSONL file path", "generated.jsonl").option("--model <model>", "Model to use for generation", "claude-sonnet-4-6").option("-j, --json", "Output JSON summary").action(async (opts) => {
     const client = new Anthropic;
     const count = parseInt(opts["count"] ?? "10");
     let seedText = "";
@@ -30746,11 +30770,22 @@ Output ${count} JSONL lines starting with {"id":"gen-001",...}`;
           valid.push(parsed);
       } catch {}
     }
+    const outputPath = opts["output"] ?? "generated.jsonl";
     const output = valid.map((c) => JSON.stringify(c)).join(`
 `);
-    writeFileSync(opts["output"] ?? "generated.jsonl", output + `
+    writeFileSync(outputPath, output + `
 `);
-    console.log(`\x1B[32m\u2713 Generated ${valid.length} cases \u2192 ${opts["output"]}\x1B[0m`);
+    if (opts["json"]) {
+      console.log(JSON.stringify({
+        generated: valid.length,
+        requested: count,
+        output: outputPath,
+        model: opts["model"] ?? "claude-sonnet-4-6",
+        description: opts["description"] ?? ""
+      }, null, 2));
+      return;
+    }
+    console.log(`\x1B[32m\u2713 Generated ${valid.length} cases \u2192 ${outputPath}\x1B[0m`);
   });
 }
 
@@ -30844,7 +30879,7 @@ function resolveApiKey(envVar, secretsPath, secretsKey) {
   return;
 }
 function doctorCommand() {
-  return new Command("doctor").description("Health check \u2014 verify API keys, DB, and config").action(async () => {
+  return new Command("doctor").description("Health check \u2014 verify API keys, DB, and config").option("-j, --json", "Output JSON").action(async (opts) => {
     const checks = [];
     const anthropicKey = resolveApiKey("ANTHROPIC_API_KEY", "hasnaxyz/anthropic/live.env", "HASNAXYZ_ANTHROPIC_LIVE_API_KEY");
     checks.push({
@@ -30888,6 +30923,15 @@ function doctorCommand() {
     } catch {
       checks.push({ name: "Example dataset (optional)", ok: false, hint: "datasets/examples/smoke.jsonl not found \u2014 install @hasna/evals globally to include examples" });
     }
+    const allOk = checks.every((c) => c.ok || c.name.toLowerCase().includes("optional"));
+    if (opts.json) {
+      console.log(JSON.stringify({
+        ok: allOk,
+        checks,
+        summary: allOk ? "All checks passed." : "Some checks failed \u2014 see hints above."
+      }, null, 2));
+      process.exit(allOk ? 0 : 1);
+    }
     console.log(`
 \x1B[1mevals doctor\x1B[0m
 `);
@@ -30896,7 +30940,6 @@ function doctorCommand() {
       console.log(`  ${icon} ${c.name}${!c.ok && c.hint ? `
       hint: ${c.hint}` : ""}`);
     }
-    const allOk = checks.every((c) => c.ok || c.name.toLowerCase().includes("optional"));
     console.log(allOk ? `
 \x1B[32m  All checks passed.\x1B[0m
 ` : `

package/dist/mcp/http.d.ts

@@ -0,0 +1,12 @@
+import { DEFAULT_MCP_HTTP_PORT, MCP_NAME } from "./server.js";
+export { DEFAULT_MCP_HTTP_PORT, MCP_NAME };
+export declare function isHttpMode(argv?: string[]): boolean;
+export declare function resolveHttpPort(argv?: string[]): number;
+export declare function handleMcpHttpRequest(req: Request): Promise<Response>;
+export interface StartHttpServerOptions {
+    port?: number;
+    hostname?: string;
+    log?: (message: string) => void;
+}
+export declare function startHttpServer(options?: StartHttpServerOptions): ReturnType<typeof Bun.serve>;
+//# sourceMappingURL=http.d.ts.map

package/dist/mcp/http.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"http.d.ts","sourceRoot":"","sources":["../../src/mcp/http.ts"],"names":[],"mappings":"AACA,OAAO,EAAe,qBAAqB,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAE3E,OAAO,EAAE,qBAAqB,EAAE,QAAQ,EAAE,CAAC;AAE3C,wBAAgB,UAAU,CAAC,IAAI,GAAE,MAAM,EAA0B,GAAG,OAAO,CAE1E;AAED,wBAAgB,eAAe,CAAC,IAAI,GAAE,MAAM,EAA0B,GAAG,MAAM,CAa9E;AAUD,wBAAsB,oBAAoB,CAAC,GAAG,EAAE,OAAO,GAAG,OAAO,CAAC,QAAQ,CAAC,CAiB1E;AAED,MAAM,WAAW,sBAAsB;IACrC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;CACjC;AAED,wBAAgB,eAAe,CAAC,OAAO,GAAE,sBAA2B,GAAG,UAAU,CAAC,OAAO,GAAG,CAAC,KAAK,CAAC,CAclG"}

package/dist/mcp/http.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=http.test.d.ts.map

package/dist/mcp/http.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"http.test.d.ts","sourceRoot":"","sources":["../../src/mcp/http.test.ts"],"names":[],"mappings":""}

package/dist/mcp/index.js

@@ -10587,8 +10587,91 @@ var _db = null;
 var init_store = () => {};
 
 // src/mcp/index.ts
-import { Server } from "@modelcontextprotocol/sdk/server/index.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
+// package.json
+var package_default = {
+  name: "@hasna/evals",
+  version: "0.1.22",
+  description: "Open source AI evaluation framework \u2014 LLM-as-judge + assertion-based evals for any AI app. CLI + MCP server.",
+  type: "module",
+  main: "dist/index.js",
+  types: "dist/index.d.ts",
+  bin: {
+    evals: "dist/cli/index.js",
+    "evals-mcp": "dist/mcp/index.js",
+    "evals-serve": "dist/server/index.js"
+  },
+  exports: {
+    ".": {
+      types: "./dist/index.d.ts",
+      import: "./dist/index.js"
+    }
+  },
+  files: [
+    "dist",
+    "dashboard/dist",
+    "datasets/examples",
+    "LICENSE",
+    "README.md"
+  ],
+  scripts: {
+    build: "cd dashboard && bun run build && cd .. && bun build src/cli/index.ts --outdir dist/cli --target bun --external ink --external react --external chalk --external @modelcontextprotocol/sdk && bun build src/mcp/index.ts --outdir dist/mcp --target bun --external @modelcontextprotocol/sdk && bun build src/server/index.ts --outdir dist/server --target bun && bun build src/index.ts --outdir dist --target bun && tsc --emitDeclarationOnly --outDir dist",
+    "build:dashboard": "cd dashboard && bun run build",
+    typecheck: "tsc --noEmit",
+    test: "bun test",
+    "dev:cli": "bun run src/cli/index.ts",
+    "dev:mcp": "bun run src/mcp/index.ts",
+    "dev:serve": "bun run src/server/index.ts",
+    prepublishOnly: "bun run typecheck && bun test && bun run build",
+    postinstall: "mkdir -p $HOME/.hasna/evals 2>/dev/null || true"
+  },
+  keywords: [
+    "evals",
+    "llm",
+    "ai",
+    "testing",
+    "evaluation",
+    "mcp",
+    "claude",
+    "llm-as-judge",
+    "typescript",
+    "cli"
+  ],
+  publishConfig: {
+    registry: "https://registry.npmjs.org",
+    access: "public"
+  },
+  repository: {
+    type: "git",
+    url: "https://github.com/hasna/open-evals.git"
+  },
+  homepage: "https://github.com/hasna/open-evals",
+  bugs: {
+    url: "https://github.com/hasna/open-evals/issues"
+  },
+  engines: {
+    bun: ">=1.0.0"
+  },
+  author: "Andrei Hasna <andrei@hasna.com>",
+  license: "Apache-2.0",
+  dependencies: {
+    "@anthropic-ai/sdk": "^0.82.0",
+    "@hasna/cloud": "^0.1.30",
+    "@modelcontextprotocol/sdk": "^1.29.0",
+    ajv: "^8.18.0",
+    chalk: "^5.4.1",
+    commander: "^14.0.3",
+    openai: "^6.33.0",
+    zod: "^4.3.6"
+  },
+  devDependencies: {
+    "@types/bun": "^1.2.4",
+    typescript: "^6.0.2"
+  }
+};
+
+// src/mcp/server.ts
+import { Server } from "@modelcontextprotocol/sdk/server/index.js";
 import { CallToolRequestSchema, ListToolsRequestSchema } from "@modelcontextprotocol/sdk/types.js";
 
 // node_modules/zod/v4/classic/external.js
@@ -30881,284 +30964,383 @@ function compareRuns(before, after) {
   };
 }
 
-// src/mcp/index.ts
+// src/mcp/server.ts
 init_store();
 import { writeFileSync, appendFileSync } from "fs";
 var pkg = await Bun.file(new URL("../../package.json", import.meta.url)).json();
-var server = new Server({ name: "evals", version: pkg.version }, { capabilities: { tools: {} } });
-var AdapterSchema = exports_external.object({
-  type: exports_external.enum(["http", "anthropic", "openai", "mcp", "function", "cli"]),
-  url: exports_external.string().optional(),
-  model: exports_external.string().optional(),
-  systemPrompt: exports_external.string().optional(),
-  command: exports_external.array(exports_external.string()).optional(),
-  tool: exports_external.string().optional(),
-  modulePath: exports_external.string().optional()
-}).passthrough();
-var tools = [
-  {
-    name: "evals_run",
-    description: "Run a full eval dataset against an app and return results",
-    inputSchema: {
-      type: "object",
-      properties: {
-        dataset: { type: "string", description: "Path to JSONL/JSON dataset" },
-        adapter: { type: "object", description: "Adapter config (type, url/model/command, etc.)" },
-        concurrency: { type: "number", description: "Parallel execution limit (default: 5)" },
-        skip_judge: { type: "boolean", description: "Skip LLM judge, run assertions only" },
-        tags: { type: "array", items: { type: "string" }, description: "Filter cases by tags" },
-        save: { type: "boolean", description: "Save run to database" },
-        output_format: { type: "string", enum: ["json", "markdown", "summary"], description: "Output format" }
-      },
-      required: ["dataset", "adapter"]
-    }
-  },
-  {
-    name: "evals_run_single",
-    description: "Run a single eval case ad-hoc \u2014 useful for agents to verify their own output quality",
-    inputSchema: {
-      type: "object",
-      properties: {
-        input: { type: "string", description: "Input to the AI app" },
-        output: { type: "string", description: "AI's response to evaluate" },
-        rubric: { type: "string", description: "Plain-English grading criteria" },
-        expected: { type: "string", description: "Expected behavior description" },
-        assertions: { type: "array", description: "Optional deterministic assertions" },
-        judge_model: { type: "string", description: "Judge model (default: claude-sonnet-4-6)" },
-        judge_provider: { type: "string", enum: ["anthropic", "openai"] }
-      },
-      required: ["input", "output", "rubric"]
-    }
-  },
-  {
-    name: "evals_judge",
-    description: "One-shot LLM judge \u2014 no dataset needed",
-    inputSchema: {
-      type: "object",
-      properties: {
-        input: { type: "string" },
-        output: { type: "string" },
-        rubric: { type: "string" },
-        expected: { type: "string" },
-        model: { type: "string" },
-        provider: { type: "string", enum: ["anthropic", "openai"] }
-      },
-      required: ["input", "output", "rubric"]
-    }
-  },
-  {
-    name: "evals_list_datasets",
-    description: "List available JSONL datasets in a directory",
-    inputSchema: {
-      type: "object",
-      properties: {
-        directory: { type: "string", description: "Directory to search (default: ./datasets)" }
+var MCP_NAME = "evals";
+var DEFAULT_MCP_HTTP_PORT = 8817;
+function buildServer() {
+  const server = new Server({ name: MCP_NAME, version: pkg.version }, { capabilities: { tools: {} } });
+  const AdapterSchema = exports_external.object({
+    type: exports_external.enum(["http", "anthropic", "openai", "mcp", "function", "cli"]),
+    url: exports_external.string().optional(),
+    model: exports_external.string().optional(),
+    systemPrompt: exports_external.string().optional(),
+    command: exports_external.array(exports_external.string()).optional(),
+    tool: exports_external.string().optional(),
+    modulePath: exports_external.string().optional()
+  }).passthrough();
+  const tools = [
+    {
+      name: "evals_run",
+      description: "Run a full eval dataset against an app and return results",
+      inputSchema: {
+        type: "object",
+        properties: {
+          dataset: { type: "string", description: "Path to JSONL/JSON dataset" },
+          adapter: { type: "object", description: "Adapter config (type, url/model/command, etc.)" },
+          concurrency: { type: "number", description: "Parallel execution limit (default: 5)" },
+          skip_judge: { type: "boolean", description: "Skip LLM judge, run assertions only" },
+          tags: { type: "array", items: { type: "string" }, description: "Filter cases by tags" },
+          save: { type: "boolean", description: "Save run to database" },
+          output_format: { type: "string", enum: ["json", "markdown", "summary"], description: "Output format" }
+        },
+        required: ["dataset", "adapter"]
       }
-    }
-  },
-  {
-    name: "evals_get_results",
-    description: "Get results for a past eval run",
-    inputSchema: {
-      type: "object",
-      properties: {
-        run_id: { type: "string", description: "Run ID or partial ID" },
-        format: { type: "string", enum: ["json", "markdown", "summary"] },
-        limit: { type: "number", description: "Max runs to list if no run_id given" }
+    },
+    {
+      name: "evals_run_single",
+      description: "Run a single eval case ad-hoc \u2014 useful for agents to verify their own output quality",
+      inputSchema: {
+        type: "object",
+        properties: {
+          input: { type: "string", description: "Input to the AI app" },
+          output: { type: "string", description: "AI's response to evaluate" },
+          rubric: { type: "string", description: "Plain-English grading criteria" },
+          expected: { type: "string", description: "Expected behavior description" },
+          assertions: { type: "array", description: "Optional deterministic assertions" },
+          judge_model: { type: "string", description: "Judge model (default: claude-sonnet-4-6)" },
+          judge_provider: { type: "string", enum: ["anthropic", "openai"] }
+        },
+        required: ["input", "output", "rubric"]
+      }
+    },
+    {
+      name: "evals_judge",
+      description: "One-shot LLM judge \u2014 no dataset needed",
+      inputSchema: {
+        type: "object",
+        properties: {
+          input: { type: "string" },
+          output: { type: "string" },
+          rubric: { type: "string" },
+          expected: { type: "string" },
+          model: { type: "string" },
+          provider: { type: "string", enum: ["anthropic", "openai"] }
+        },
+        required: ["input", "output", "rubric"]
+      }
+    },
+    {
+      name: "evals_list_datasets",
+      description: "List available JSONL datasets in a directory",
+      inputSchema: {
+        type: "object",
+        properties: {
+          directory: { type: "string", description: "Directory to search (default: ./datasets)" }
+        }
+      }
+    },
+    {
+      name: "evals_get_results",
+      description: "Get results for a past eval run",
+      inputSchema: {
+        type: "object",
+        properties: {
+          run_id: { type: "string", description: "Run ID or partial ID" },
+          format: { type: "string", enum: ["json", "markdown", "summary"] },
+          limit: { type: "number", description: "Max runs to list if no run_id given" }
+        }
+      }
+    },
+    {
+      name: "evals_compare",
+      description: "Compare two eval runs \u2014 show regressions and improvements",
+      inputSchema: {
+        type: "object",
+        properties: {
+          before: { type: "string", description: "Before run ID or baseline name" },
+          after: { type: "string", description: "After run ID" }
+        },
+        required: ["before", "after"]
+      }
+    },
+    {
+      name: "evals_create_case",
+      description: "Add a new eval case to a dataset file",
+      inputSchema: {
+        type: "object",
+        properties: {
+          dataset: { type: "string", description: "Path to JSONL file to append to" },
+          id: { type: "string", description: "Unique case ID" },
+          input: { type: "string" },
+          expected: { type: "string" },
+          rubric: { type: "string", description: "Judge rubric for this case" },
+          assertions: { type: "array" },
+          tags: { type: "array", items: { type: "string" } }
+        },
+        required: ["dataset", "id", "input"]
+      }
+    },
+    {
+      name: "evals_generate_cases",
+      description: "Auto-generate eval cases from a description using Claude",
+      inputSchema: {
+        type: "object",
+        properties: {
+          description: { type: "string", description: "What behavior to test" },
+          count: { type: "number", description: "Number of cases to generate (default: 10)" },
+          output: { type: "string", description: "Output JSONL path" },
+          seeds: { type: "string", description: "Path to seed examples JSONL" }
+        },
+        required: ["description"]
       }
     }
-  },
-  {
-    name: "evals_compare",
-    description: "Compare two eval runs \u2014 show regressions and improvements",
-    inputSchema: {
-      type: "object",
-      properties: {
-        before: { type: "string", description: "Before run ID or baseline name" },
-        after: { type: "string", description: "After run ID" }
-      },
-      required: ["before", "after"]
-    }
-  },
-  {
-    name: "evals_create_case",
-    description: "Add a new eval case to a dataset file",
-    inputSchema: {
-      type: "object",
-      properties: {
-        dataset: { type: "string", description: "Path to JSONL file to append to" },
-        id: { type: "string", description: "Unique case ID" },
-        input: { type: "string" },
-        expected: { type: "string" },
-        rubric: { type: "string", description: "Judge rubric for this case" },
-        assertions: { type: "array" },
-        tags: { type: "array", items: { type: "string" } }
-      },
-      required: ["dataset", "id", "input"]
-    }
-  },
-  {
-    name: "evals_generate_cases",
-    description: "Auto-generate eval cases from a description using Claude",
-    inputSchema: {
-      type: "object",
-      properties: {
-        description: { type: "string", description: "What behavior to test" },
-        count: { type: "number", description: "Number of cases to generate (default: 10)" },
-        output: { type: "string", description: "Output JSONL path" },
-        seeds: { type: "string", description: "Path to seed examples JSONL" }
-      },
-      required: ["description"]
-    }
-  }
-];
-server.setRequestHandler(ListToolsRequestSchema, async () => ({ tools }));
-server.setRequestHandler(CallToolRequestSchema, async (req) => {
-  const { name, arguments: args } = req.params;
-  const a = args ?? {};
-  try {
-    switch (name) {
-      case "evals_run": {
-        const adapter = AdapterSchema.parse(a["adapter"]);
-        const { cases } = await loadDataset(String(a["dataset"]), {
-          tags: a["tags"]
-        });
-        const run = await runEvals(cases, {
-          dataset: String(a["dataset"]),
-          adapter,
-          concurrency: Number(a["concurrency"] ?? 5),
-          skipJudge: Boolean(a["skip_judge"])
-        });
-        if (a["save"])
-          saveRun(run);
-        const fmt = String(a["output_format"] ?? "summary");
-        const output = fmt === "json" ? toJson(run) : fmt === "markdown" ? toMarkdown(run) : `${run.stats.passed}/${run.stats.total} passed (${(run.stats.passRate * 100).toFixed(1)}%) in ${run.stats.totalDurationMs}ms. Run ID: ${run.id.slice(0, 8)}`;
-        return { content: [{ type: "text", text: output }] };
-      }
-      case "evals_run_single": {
-        const evalCase = {
-          id: "mcp-single",
-          input: String(a["input"]),
-          expected: a["expected"] ? String(a["expected"]) : undefined,
-          assertions: a["assertions"] ?? [],
-          judge: {
+  ];
+  server.setRequestHandler(ListToolsRequestSchema, async () => ({ tools }));
+  server.setRequestHandler(CallToolRequestSchema, async (req) => {
+    const { name, arguments: args } = req.params;
+    const a = args ?? {};
+    try {
+      switch (name) {
+        case "evals_run": {
+          const adapter = AdapterSchema.parse(a["adapter"]);
+          const { cases } = await loadDataset(String(a["dataset"]), {
+            tags: a["tags"]
+          });
+          const run = await runEvals(cases, {
+            dataset: String(a["dataset"]),
+            adapter,
+            concurrency: Number(a["concurrency"] ?? 5),
+            skipJudge: Boolean(a["skip_judge"])
+          });
+          if (a["save"])
+            saveRun(run);
+          const fmt = String(a["output_format"] ?? "summary");
+          const output = fmt === "json" ? toJson(run) : fmt === "markdown" ? toMarkdown(run) : `${run.stats.passed}/${run.stats.total} passed (${(run.stats.passRate * 100).toFixed(1)}%) in ${run.stats.totalDurationMs}ms. Run ID: ${run.id.slice(0, 8)}`;
+          return { content: [{ type: "text", text: output }] };
+        }
+        case "evals_run_single": {
+          const evalCase = {
+            id: "mcp-single",
+            input: String(a["input"]),
+            expected: a["expected"] ? String(a["expected"]) : undefined,
+            assertions: a["assertions"] ?? [],
+            judge: {
+              rubric: String(a["rubric"]),
+              model: a["judge_model"] ? String(a["judge_model"]) : undefined,
+              provider: a["judge_provider"]
+            }
+          };
+          const mockAdapter = { type: "function", modulePath: "__mock__" };
+          const judgeResult = await judgeOnce({
+            input: String(a["input"]),
+            output: String(a["output"]),
             rubric: String(a["rubric"]),
+            expected: a["expected"] ? String(a["expected"]) : undefined,
             model: a["judge_model"] ? String(a["judge_model"]) : undefined,
             provider: a["judge_provider"]
-          }
-        };
-        const mockAdapter = { type: "function", modulePath: "__mock__" };
-        const judgeResult = await judgeOnce({
-          input: String(a["input"]),
-          output: String(a["output"]),
-          rubric: String(a["rubric"]),
-          expected: a["expected"] ? String(a["expected"]) : undefined,
-          model: a["judge_model"] ? String(a["judge_model"]) : undefined,
-          provider: a["judge_provider"]
-        });
-        return {
-          content: [{
-            type: "text",
-            text: `VERDICT: ${judgeResult.verdict}
+          });
+          return {
+            content: [{
+              type: "text",
+              text: `VERDICT: ${judgeResult.verdict}
 REASONING: ${judgeResult.reasoning}`
-          }]
-        };
-      }
-      case "evals_judge": {
-        const r = await judgeOnce({
-          input: String(a["input"]),
-          output: String(a["output"]),
-          rubric: String(a["rubric"]),
-          expected: a["expected"] ? String(a["expected"]) : undefined,
-          model: a["model"] ? String(a["model"]) : undefined,
-          provider: a["provider"]
-        });
-        return { content: [{ type: "text", text: `${r.verdict}
+            }]
+          };
+        }
+        case "evals_judge": {
+          const r = await judgeOnce({
+            input: String(a["input"]),
+            output: String(a["output"]),
+            rubric: String(a["rubric"]),
+            expected: a["expected"] ? String(a["expected"]) : undefined,
+            model: a["model"] ? String(a["model"]) : undefined,
+            provider: a["provider"]
+          });
+          return { content: [{ type: "text", text: `${r.verdict}
 ${r.reasoning}` }] };
-      }
-      case "evals_list_datasets": {
-        const dir = String(a["directory"] ?? "./datasets");
-        const files = [];
-        for await (const f of new Bun.Glob(`${dir}/**/*.jsonl`).scan("."))
-          files.push(f);
-        for await (const f of new Bun.Glob(`${dir}/**/*.json`).scan("."))
-          files.push(f);
-        return { content: [{ type: "text", text: files.length > 0 ? files.join(`
+        }
+        case "evals_list_datasets": {
+          const dir = String(a["directory"] ?? "./datasets");
+          const files = [];
+          for await (const f of new Bun.Glob(`${dir}/**/*.jsonl`).scan("."))
+            files.push(f);
+          for await (const f of new Bun.Glob(`${dir}/**/*.json`).scan("."))
+            files.push(f);
+          return { content: [{ type: "text", text: files.length > 0 ? files.join(`
 `) : "No datasets found" }] };
-      }
-      case "evals_get_results": {
-        if (a["run_id"]) {
-          const run = getRun(String(a["run_id"]));
-          if (!run)
-            return { content: [{ type: "text", text: "Run not found" }] };
-          const fmt = String(a["format"] ?? "summary");
-          const text = fmt === "json" ? toJson(run) : fmt === "markdown" ? toMarkdown(run) : `Run ${run.id.slice(0, 8)}: ${run.stats.passed}/${run.stats.total} passed (${(run.stats.passRate * 100).toFixed(1)}%)`;
-          return { content: [{ type: "text", text }] };
-        } else {
-          const runs = listRuns(Number(a["limit"] ?? 10));
-          const summary = runs.map((r) => `${r.id.slice(0, 8)} | ${r.createdAt.slice(0, 10)} | ${r.dataset} | ${r.stats.passed}/${r.stats.total} passed`).join(`
+        }
+        case "evals_get_results": {
+          if (a["run_id"]) {
+            const run = getRun(String(a["run_id"]));
+            if (!run)
+              return { content: [{ type: "text", text: "Run not found" }] };
+            const fmt = String(a["format"] ?? "summary");
+            const text = fmt === "json" ? toJson(run) : fmt === "markdown" ? toMarkdown(run) : `Run ${run.id.slice(0, 8)}: ${run.stats.passed}/${run.stats.total} passed (${(run.stats.passRate * 100).toFixed(1)}%)`;
+            return { content: [{ type: "text", text }] };
+          } else {
+            const runs = listRuns(Number(a["limit"] ?? 10));
+            const summary = runs.map((r) => `${r.id.slice(0, 8)} | ${r.createdAt.slice(0, 10)} | ${r.dataset} | ${r.stats.passed}/${r.stats.total} passed`).join(`
 `);
-          return { content: [{ type: "text", text: summary || "No runs found" }] };
-        }
-      }
-      case "evals_compare": {
-        const { getBaseline: getBaseline2 } = await Promise.resolve().then(() => (init_store(), exports_store));
-        const before = getRun(String(a["before"])) ?? getBaseline2(String(a["before"]));
-        const after = getRun(String(a["after"])) ?? getBaseline2(String(a["after"]));
-        if (!before || !after)
-          return { content: [{ type: "text", text: "Run(s) not found" }] };
-        const diff = compareRuns(before, after);
-        const lines = [
-          `Score delta: ${diff.passRateDelta >= 0 ? "+" : ""}${(diff.passRateDelta * 100).toFixed(1)}%`,
-          ...diff.regressions.map((r) => `\u2193 REGRESSION: ${r.caseId} (${r.before} \u2192 ${r.after})`),
-          ...diff.improvements.map((i) => `\u2191 IMPROVEMENT: ${i.caseId} (${i.before} \u2192 ${i.after})`)
-        ];
-        return { content: [{ type: "text", text: lines.join(`
+            return { content: [{ type: "text", text: summary || "No runs found" }] };
+          }
+        }
+        case "evals_compare": {
+          const { getBaseline: getBaseline2 } = await Promise.resolve().then(() => (init_store(), exports_store));
+          const before = getRun(String(a["before"])) ?? getBaseline2(String(a["before"]));
+          const after = getRun(String(a["after"])) ?? getBaseline2(String(a["after"]));
+          if (!before || !after)
+            return { content: [{ type: "text", text: "Run(s) not found" }] };
+          const diff = compareRuns(before, after);
+          const lines = [
+            `Score delta: ${diff.passRateDelta >= 0 ? "+" : ""}${(diff.passRateDelta * 100).toFixed(1)}%`,
+            ...diff.regressions.map((r) => `\u2193 REGRESSION: ${r.caseId} (${r.before} \u2192 ${r.after})`),
+            ...diff.improvements.map((i) => `\u2191 IMPROVEMENT: ${i.caseId} (${i.before} \u2192 ${i.after})`)
+          ];
+          return { content: [{ type: "text", text: lines.join(`
 `) }] };
-      }
-      case "evals_create_case": {
-        const evalCase = {
-          id: String(a["id"]),
-          input: String(a["input"]),
-          expected: a["expected"] ? String(a["expected"]) : undefined,
-          judge: a["rubric"] ? { rubric: String(a["rubric"]) } : undefined,
-          assertions: a["assertions"] ?? undefined,
-          tags: a["tags"] ?? undefined
-        };
-        appendFileSync(String(a["dataset"]), JSON.stringify(evalCase) + `
+        }
+        case "evals_create_case": {
+          const evalCase = {
+            id: String(a["id"]),
+            input: String(a["input"]),
+            expected: a["expected"] ? String(a["expected"]) : undefined,
+            judge: a["rubric"] ? { rubric: String(a["rubric"]) } : undefined,
+            assertions: a["assertions"] ?? undefined,
+            tags: a["tags"] ?? undefined
+          };
+          appendFileSync(String(a["dataset"]), JSON.stringify(evalCase) + `
 `);
-        return { content: [{ type: "text", text: `Case "${evalCase.id}" appended to ${a["dataset"]}` }] };
-      }
-      case "evals_generate_cases": {
-        const Anthropic2 = (await Promise.resolve().then(() => (init_sdk(), exports_sdk))).default;
-        const client = new Anthropic2;
-        const count = Number(a["count"] ?? 10);
-        const res = await client.messages.create({
-          model: "claude-sonnet-4-6",
-          max_tokens: 4096,
-          temperature: 1,
-          system: "Generate eval cases as JSONL. Each line: {id, input, expected, judge: {rubric}, tags}. Output only JSONL lines.",
-          messages: [{ role: "user", content: `Generate ${count} eval cases for: ${a["description"]}` }]
-        });
-        const text = res.content.filter((b) => b.type === "text").map((b) => b.text).join(`
+          return { content: [{ type: "text", text: `Case "${evalCase.id}" appended to ${a["dataset"]}` }] };
+        }
+        case "evals_generate_cases": {
+          const Anthropic2 = (await Promise.resolve().then(() => (init_sdk(), exports_sdk))).default;
+          const client = new Anthropic2;
+          const count = Number(a["count"] ?? 10);
+          const res = await client.messages.create({
+            model: "claude-sonnet-4-6",
+            max_tokens: 4096,
+            temperature: 1,
+            system: "Generate eval cases as JSONL. Each line: {id, input, expected, judge: {rubric}, tags}. Output only JSONL lines.",
+            messages: [{ role: "user", content: `Generate ${count} eval cases for: ${a["description"]}` }]
+          });
+          const text = res.content.filter((b) => b.type === "text").map((b) => b.text).join(`
 `);
-        const lines = text.split(`
+          const lines = text.split(`
 `).filter((l) => l.trim().startsWith("{"));
-        const output = String(a["output"] ?? "generated.jsonl");
-        writeFileSync(output, lines.join(`
+          const output = String(a["output"] ?? "generated.jsonl");
+          writeFileSync(output, lines.join(`
 `) + `
 `);
-        return { content: [{ type: "text", text: `Generated ${lines.length} cases \u2192 ${output}` }] };
+          return { content: [{ type: "text", text: `Generated ${lines.length} cases \u2192 ${output}` }] };
+        }
+        default:
+          return { content: [{ type: "text", text: `Unknown tool: ${name}` }] };
       }
-      default:
-        return { content: [{ type: "text", text: `Unknown tool: ${name}` }] };
+    } catch (err) {
+      return {
+        content: [{ type: "text", text: `Error: ${err instanceof Error ? err.message : String(err)}` }],
+        isError: true
+      };
     }
-  } catch (err) {
-    return {
-      content: [{ type: "text", text: `Error: ${err instanceof Error ? err.message : String(err)}` }],
-      isError: true
-    };
+  });
+  return server;
+}
+
+// src/mcp/http.ts
+import { WebStandardStreamableHTTPServerTransport } from "@modelcontextprotocol/sdk/server/webStandardStreamableHttp.js";
+function isHttpMode(argv = process.argv.slice(2)) {
+  return argv.includes("--http") || process.env["MCP_HTTP"] === "1";
+}
+function resolveHttpPort(argv = process.argv.slice(2)) {
+  for (let i = 0;i < argv.length; i++) {
+    const arg = argv[i];
+    if (arg === "--port" || arg === "-p") {
+      const raw = argv[i + 1];
+      if (!raw)
+        throw new Error(`Invalid port: ${raw ?? ""}`);
+      return parsePort(raw, "port");
+    }
+  }
+  const fromEnv = process.env["MCP_HTTP_PORT"];
+  if (fromEnv)
+    return parsePort(fromEnv, "MCP_HTTP_PORT");
+  return DEFAULT_MCP_HTTP_PORT;
+}
+function parsePort(raw, label) {
+  const value = Number(raw);
+  if (!Number.isInteger(value) || value < 1 || value > 65535) {
+    throw new Error(`Invalid ${label}: ${raw}`);
+  }
+  return value;
+}
+async function handleMcpHttpRequest(req) {
+  const url2 = new URL(req.url);
+  if (url2.pathname === "/health" && req.method === "GET") {
+    return Response.json({ status: "ok", name: MCP_NAME });
   }
+  if (url2.pathname === "/mcp") {
+    const transport = new WebStandardStreamableHTTPServerTransport({
+      sessionIdGenerator: undefined
+    });
+    const server = buildServer();
+    await server.connect(transport);
+    return transport.handleRequest(req);
+  }
+  return new Response("Not Found", { status: 404 });
+}
+function startHttpServer(options = {}) {
+  const port = options.port ?? DEFAULT_MCP_HTTP_PORT;
+  const hostname3 = options.hostname ?? "127.0.0.1";
+  const log2 = options.log ?? console.error;
+  const server = Bun.serve({
+    port,
+    hostname: hostname3,
+    fetch: handleMcpHttpRequest
+  });
+  const address = `http://${hostname3}:${server.port}`;
+  log2(`${MCP_NAME}-mcp HTTP listening on ${address}/mcp (health: ${address}/health)`);
+  return server;
+}
+
+// src/mcp/index.ts
+function printHelp() {
+  console.log(`Usage: evals-mcp [options]
+
+Runs the @hasna/evals MCP server (stdio by default).
+
+Options:
+      --http         Serve MCP over Streamable HTTP on 127.0.0.1
+  -p, --port <port>  HTTP port (default: MCP_HTTP_PORT or 8817)
+  -V, --version      output the version number
+  -h, --help         display help for command
+
+Environment:
+  MCP_HTTP=1         Enable HTTP mode
+  MCP_HTTP_PORT      Override default HTTP port (8817)`);
+}
+var args = process.argv.slice(2);
+if (args.includes("--help") || args.includes("-h")) {
+  printHelp();
+  process.exit(0);
+}
+if (args.includes("--version") || args.includes("-V")) {
+  console.log(package_default.version);
+  process.exit(0);
+}
+async function main() {
+  if (isHttpMode(args)) {
+    startHttpServer({ port: resolveHttpPort(args) });
+    await new Promise(() => {});
+    return;
+  }
+  const server = buildServer();
+  const transport = new StdioServerTransport;
+  await server.connect(transport);
+}
+main().catch((error50) => {
+  console.error("MCP server error:", error50);
+  process.exit(1);
 });
-var transport = new StdioServerTransport;
-await server.connect(transport);

package/dist/mcp/server.d.ts

@@ -0,0 +1,5 @@
+import { Server } from "@modelcontextprotocol/sdk/server/index.js";
+export declare const MCP_NAME = "evals";
+export declare const DEFAULT_MCP_HTTP_PORT = 8817;
+export declare function buildServer(): Server;
+//# sourceMappingURL=server.d.ts.map

package/dist/mcp/server.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/mcp/server.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AAanE,eAAO,MAAM,QAAQ,UAAU,CAAC;AAChC,eAAO,MAAM,qBAAqB,OAAO,CAAC;AAE1C,wBAAgB,WAAW,IAAI,MAAM,CAmSpC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hasna/evals",
-  "version": "0.1.20",
+  "version": "0.1.22",
   "description": "Open source AI evaluation framework — LLM-as-judge + assertion-based evals for any AI app. CLI + MCP server.",
   "type": "module",
   "main": "dist/index.js",