@langwatch/mcp-server 2.0.0 → 2.2.0

package/dist/update-prompt-GDKRYJI3.js

@@ -0,0 +1,140 @@
+import {
+  getPrompt,
+  getPromptVersions,
+  updatePrompt
+} from "./chunk-UWLE5T2F.js";
+import "./chunk-H6WGMG2T.js";
+
+// src/tools/update-prompt.ts
+async function handleUpdatePrompt(params) {
+  const { idOrHandle, ...data } = params;
+  let updated;
+  try {
+    updated = await updatePrompt(idOrHandle, data);
+  } catch (err) {
+    if (params.tags && params.tags.length > 0) {
+      return renderTagFailure({ idOrHandle, params });
+    }
+    throw err;
+  }
+  return renderUpdateSuccess({ idOrHandle, params, updated });
+}
+async function identifyNewVersion({
+  idOrHandle,
+  prompt,
+  commitMessage
+}) {
+  if (prompt.commitMessage === commitMessage) {
+    return { version: prompt.version, versionId: prompt.versionId };
+  }
+  try {
+    const versions = await getPromptVersions(idOrHandle);
+    const match = (Array.isArray(versions) ? versions : []).find(
+      (v) => v.commitMessage === commitMessage
+    );
+    if (match) return { version: match.version, versionId: match.versionId };
+  } catch {
+  }
+  return void 0;
+}
+function deploymentTagsOf(prompt) {
+  return (prompt.tags ?? []).filter((tag) => tag?.name !== "latest");
+}
+function deploymentLines({
+  prompt,
+  newVersion
+}) {
+  const allDeployments = deploymentTagsOf(prompt);
+  const newTags = allDeployments.filter((tag) => newVersion.versionId != null && tag.versionId === newVersion.versionId).map((tag) => tag.name);
+  if (newTags.length > 0) {
+    return [`**Deployed to**: ${newTags.join(", ")}`];
+  }
+  const lines = [`**Deployment**: not deployed`];
+  const otherTags = Array.from(
+    new Set(
+      allDeployments.filter((tag) => tag.versionId !== newVersion.versionId).map((tag) => tag.name)
+    )
+  );
+  if (otherTags.length > 0) {
+    lines.push(`**Existing deployments (untouched)**: ${otherTags.join(", ")}`);
+  }
+  return lines;
+}
+async function renderUpdateSuccess({
+  idOrHandle,
+  params,
+  updated
+}) {
+  let prompt;
+  try {
+    prompt = await getPrompt(idOrHandle, {
+      version: void 0,
+      tag: void 0
+    });
+  } catch {
+    const lines2 = [];
+    lines2.push("Prompt updated successfully!\n");
+    if (updated?.id) lines2.push(`**ID**: ${updated.id}`);
+    if (updated?.handle) lines2.push(`**Handle**: ${updated.handle}`);
+    lines2.push(`**Commit**: ${params.commitMessage}`);
+    lines2.push(
+      `**Note**: the update succeeded, but version and deployment details are unavailable (confirmation read failed). Run platform_get_prompt to inspect the current state. Do not retry the update.`
+    );
+    return lines2.join("\n");
+  }
+  const newVersion = await identifyNewVersion({
+    idOrHandle,
+    prompt,
+    commitMessage: params.commitMessage
+  });
+  const lines = [];
+  lines.push("Prompt updated successfully!\n");
+  if (prompt.id) lines.push(`**ID**: ${prompt.id}`);
+  if (prompt.handle) lines.push(`**Handle**: ${prompt.handle}`);
+  if (newVersion?.version != null) lines.push(`**Version**: v${newVersion.version}`);
+  if (newVersion?.versionId) lines.push(`**Version ID**: ${newVersion.versionId}`);
+  lines.push(`**Commit**: ${params.commitMessage}`);
+  if (!newVersion) {
+    lines.push(
+      `**Note**: update succeeded, but the new version could not be identified in the re-fetched prompt \u2014 version and deployment details are unavailable. Run platform_get_prompt to inspect the current state.`
+    );
+  }
+  if (newVersion) {
+    lines.push(...deploymentLines({ prompt, newVersion }));
+  }
+  return lines.join("\n");
+}
+async function renderTagFailure({
+  idOrHandle,
+  params
+}) {
+  const failedTags = params.tags?.join(", ") ?? "";
+  let prompt;
+  try {
+    prompt = await getPrompt(idOrHandle, {
+      version: void 0,
+      tag: void 0
+    });
+  } catch {
+    return `Prompt update failed: could not assign tag(s) ${failedTags} to "${idOrHandle}". Whether a new version was created could not be confirmed (confirmation read failed) \u2014 run platform_get_prompt before retrying.`;
+  }
+  const matched = await identifyNewVersion({
+    idOrHandle,
+    prompt,
+    commitMessage: params.commitMessage
+  });
+  if (matched) {
+    const lines = [];
+    lines.push("Prompt update partially failed.\n");
+    lines.push("A new version was created, but assigning the requested tag(s) failed.");
+    if (matched.versionId) lines.push(`**Version ID**: ${matched.versionId}`);
+    lines.push(`**Status**: created, untagged`);
+    lines.push(`**Failed tag(s)**: ${failedTags}`);
+    return lines.join("\n");
+  }
+  return `Prompt update failed: could not assign tag(s) ${failedTags} to "${idOrHandle}".`;
+}
+export {
+  handleUpdatePrompt
+};
+//# sourceMappingURL=update-prompt-GDKRYJI3.js.map

package/dist/update-prompt-MY263DYF.js

@@ -0,0 +1,141 @@
+import {
+  getPrompt,
+  getPromptVersions,
+  updatePrompt
+} from "./chunk-UWLE5T2F.js";
+import "./chunk-H6WGMG2T.js";
+import "./chunk-SAS5OB7F.js";
+
+// src/tools/update-prompt.ts
+async function handleUpdatePrompt(params) {
+  const { idOrHandle, ...data } = params;
+  let updated;
+  try {
+    updated = await updatePrompt(idOrHandle, data);
+  } catch (err) {
+    if (params.tags && params.tags.length > 0) {
+      return renderTagFailure({ idOrHandle, params });
+    }
+    throw err;
+  }
+  return renderUpdateSuccess({ idOrHandle, params, updated });
+}
+async function identifyNewVersion({
+  idOrHandle,
+  prompt,
+  commitMessage
+}) {
+  if (prompt.commitMessage === commitMessage) {
+    return { version: prompt.version, versionId: prompt.versionId };
+  }
+  try {
+    const versions = await getPromptVersions(idOrHandle);
+    const match = (Array.isArray(versions) ? versions : []).find(
+      (v) => v.commitMessage === commitMessage
+    );
+    if (match) return { version: match.version, versionId: match.versionId };
+  } catch {
+  }
+  return void 0;
+}
+function deploymentTagsOf(prompt) {
+  return (prompt.tags ?? []).filter((tag) => tag?.name !== "latest");
+}
+function deploymentLines({
+  prompt,
+  newVersion
+}) {
+  const allDeployments = deploymentTagsOf(prompt);
+  const newTags = allDeployments.filter((tag) => newVersion.versionId != null && tag.versionId === newVersion.versionId).map((tag) => tag.name);
+  if (newTags.length > 0) {
+    return [`**Deployed to**: ${newTags.join(", ")}`];
+  }
+  const lines = [`**Deployment**: not deployed`];
+  const otherTags = Array.from(
+    new Set(
+      allDeployments.filter((tag) => tag.versionId !== newVersion.versionId).map((tag) => tag.name)
+    )
+  );
+  if (otherTags.length > 0) {
+    lines.push(`**Existing deployments (untouched)**: ${otherTags.join(", ")}`);
+  }
+  return lines;
+}
+async function renderUpdateSuccess({
+  idOrHandle,
+  params,
+  updated
+}) {
+  let prompt;
+  try {
+    prompt = await getPrompt(idOrHandle, {
+      version: void 0,
+      tag: void 0
+    });
+  } catch {
+    const lines2 = [];
+    lines2.push("Prompt updated successfully!\n");
+    if (updated?.id) lines2.push(`**ID**: ${updated.id}`);
+    if (updated?.handle) lines2.push(`**Handle**: ${updated.handle}`);
+    lines2.push(`**Commit**: ${params.commitMessage}`);
+    lines2.push(
+      `**Note**: the update succeeded, but version and deployment details are unavailable (confirmation read failed). Run platform_get_prompt to inspect the current state. Do not retry the update.`
+    );
+    return lines2.join("\n");
+  }
+  const newVersion = await identifyNewVersion({
+    idOrHandle,
+    prompt,
+    commitMessage: params.commitMessage
+  });
+  const lines = [];
+  lines.push("Prompt updated successfully!\n");
+  if (prompt.id) lines.push(`**ID**: ${prompt.id}`);
+  if (prompt.handle) lines.push(`**Handle**: ${prompt.handle}`);
+  if (newVersion?.version != null) lines.push(`**Version**: v${newVersion.version}`);
+  if (newVersion?.versionId) lines.push(`**Version ID**: ${newVersion.versionId}`);
+  lines.push(`**Commit**: ${params.commitMessage}`);
+  if (!newVersion) {
+    lines.push(
+      `**Note**: update succeeded, but the new version could not be identified in the re-fetched prompt \u2014 version and deployment details are unavailable. Run platform_get_prompt to inspect the current state.`
+    );
+  }
+  if (newVersion) {
+    lines.push(...deploymentLines({ prompt, newVersion }));
+  }
+  return lines.join("\n");
+}
+async function renderTagFailure({
+  idOrHandle,
+  params
+}) {
+  const failedTags = params.tags?.join(", ") ?? "";
+  let prompt;
+  try {
+    prompt = await getPrompt(idOrHandle, {
+      version: void 0,
+      tag: void 0
+    });
+  } catch {
+    return `Prompt update failed: could not assign tag(s) ${failedTags} to "${idOrHandle}". Whether a new version was created could not be confirmed (confirmation read failed) \u2014 run platform_get_prompt before retrying.`;
+  }
+  const matched = await identifyNewVersion({
+    idOrHandle,
+    prompt,
+    commitMessage: params.commitMessage
+  });
+  if (matched) {
+    const lines = [];
+    lines.push("Prompt update partially failed.\n");
+    lines.push("A new version was created, but assigning the requested tag(s) failed.");
+    if (matched.versionId) lines.push(`**Version ID**: ${matched.versionId}`);
+    lines.push(`**Status**: created, untagged`);
+    lines.push(`**Failed tag(s)**: ${failedTags}`);
+    return lines.join("\n");
+  }
+  return `Prompt update failed: could not assign tag(s) ${failedTags} to "${idOrHandle}".`;
+}
+export {
+  handleUpdatePrompt
+};
+//# sourceMappingURL=update-prompt-MY263DYF.js.map

package/dist/update-scenario-7KA3JDXA.js

@@ -0,0 +1,34 @@
+import {
+  formatScenarioFields
+} from "./chunk-XUIJYNRJ.js";
+import {
+  updateScenario
+} from "./chunk-I7ATIZED.js";
+import "./chunk-UWLE5T2F.js";
+import "./chunk-H6WGMG2T.js";
+
+// src/tools/update-scenario.ts
+async function handleUpdateScenario(params) {
+  const { scenarioId, ...data } = params;
+  const result = await updateScenario({ id: scenarioId, ...data });
+  const lines = [];
+  lines.push("Scenario updated successfully!\n");
+  lines.push(`**ID**: ${result.id}`);
+  lines.push(`**Name**: ${result.name}`);
+  if (result.situation) lines.push(`**Situation**: ${result.situation}`);
+  if (Array.isArray(result.criteria) && result.criteria.length > 0) {
+    lines.push(`**Criteria**: ${result.criteria.length} criteria`);
+  }
+  if (Array.isArray(result.labels) && result.labels.length > 0) {
+    lines.push(`**Labels**: ${result.labels.join(", ")}`);
+  }
+  if (result.testSuiteId) {
+    lines.push(`**Test suite**: ${result.testSuiteId}`);
+  }
+  lines.push(...formatScenarioFields(result.fields));
+  return lines.join("\n");
+}
+export {
+  handleUpdateScenario
+};
+//# sourceMappingURL=update-scenario-7KA3JDXA.js.map

package/dist/{update-scenario-NEXCAQI7.js → update-scenario-MOHRQKAQ.js}

@@ -1,8 +1,11 @@
+import {
+  formatScenarioFields
+} from "./chunk-XUIJYNRJ.js";
 import {
   updateScenario
-} from "./chunk-J42FWRJ2.js";
-import "./chunk-A2LEVYQ5.js";
-import "./chunk-WT4PO4A3.js";
+} from "./chunk-I7ATIZED.js";
+import "./chunk-UWLE5T2F.js";
+import "./chunk-H6WGMG2T.js";
 import "./chunk-SAS5OB7F.js";
 
 // src/tools/update-scenario.ts
@@ -20,9 +23,13 @@ async function handleUpdateScenario(params) {
   if (Array.isArray(result.labels) && result.labels.length > 0) {
     lines.push(`**Labels**: ${result.labels.join(", ")}`);
   }
+  if (result.testSuiteId) {
+    lines.push(`**Test suite**: ${result.testSuiteId}`);
+  }
+  lines.push(...formatScenarioFields(result.fields));
   return lines.join("\n");
 }
 export {
   handleUpdateScenario
 };
-//# sourceMappingURL=update-scenario-NEXCAQI7.js.map
+//# sourceMappingURL=update-scenario-MOHRQKAQ.js.map

package/package.json

@@ -1,81 +1,76 @@
 {
   "name": "@langwatch/mcp-server",
-  "version": "2.0.0",
+  "version": "2.2.0",
   "description": "An MCP server for Langwatch.",
+  "license": "MIT",
+  "author": "",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/langwatch/langwatch.git",
+    "directory": "mcp-server"
+  },
+  "bin": {
+    "langwatch-mcp-server": "./dist/index.js"
+  },
   "type": "module",
   "main": "./dist/index.js",
   "module": "./dist/index.js",
   "types": "./dist/index.d.ts",
   "exports": {
     ".": {
-      "import": "./dist/index.js",
-      "require": "./dist/index.js"
+      "import": "./dist/index.js"
     },
     "./create-mcp-server": {
       "types": "./src/create-mcp-server.d.ts",
       "import": "./dist/create-mcp-server.js",
-      "require": "./dist/create-mcp-server.js",
+      "require": "./dist/create-mcp-server.cjs",
       "default": "./dist/create-mcp-server.js"
     },
     "./config": {
       "types": "./src/config.d.ts",
       "import": "./dist/config.js",
-      "require": "./dist/config.js",
+      "require": "./dist/config.cjs",
       "default": "./dist/config.js"
     }
   },
-  "repository": {
-    "type": "git",
-    "url": "https://github.com/langwatch/langwatch.git",
-    "directory": "mcp-server"
-  },
-  "author": "",
-  "license": "MIT",
-  "engines": {
-    "node": ">=18.0.0"
+  "dependencies": {
+    "@langwatch/evaluator-contract": "0.1.0",
+    "@modelcontextprotocol/sdk": "1.29.0",
+    "zod": "^4.6.7"
   },
   "devDependencies": {
     "@ai-sdk/anthropic": "^3.0.44",
-    "@anthropic-ai/claude-code": "2.1.202",
-    "@eslint/js": "^10.0.1",
-    "@langwatch/scenario": "^0.4.6",
-    "@opentelemetry/context-async-hooks": "^2.1.0",
-    "@opentelemetry/sdk-node": "^0.219.0",
+    "@anthropic-ai/claude-code": "2.1.228",
+    "@langwatch/egress": "0.0.1",
+    "@langwatch/handled-error": "0.1.0",
+    "@langwatch/redaction": "0.0.1",
+    "@langwatch/scenario": "1.7.0-dev.voice10",
     "@types/debug": "^4.1.13",
     "@types/express": "^5.0.6",
-    "@types/node": "^25.6.0",
+    "@types/node": "^26.2.0",
     "@types/yargs": "^17.0.33",
-    "@typescript/native-preview": "7.0.0-dev.20260705.1",
-    "ai": "^6.0.161",
-    "chalk": "^5.6.2",
-    "date-fns": "^4.1.0",
-    "dotenv": "^17.2.2",
-    "esbuild": "^0.28.1",
-    "eslint": "^10.2.1",
+    "ai": "^6.0.217",
+    "chalk": "^6.0.0",
+    "dotenv": "^17.4.2",
+    "esbuild": "^0.28.2",
     "express": "^5.2.1",
     "node-pty": "^1.0.0",
-    "tsup": "^8.1.0",
-    "tsx": "^4.20.5",
-    "typescript": "^5.0.0",
-    "typescript-eslint": "^8.55.0",
-    "vitest": "^4.1.8",
+    "tsup": "^8.5.0",
+    "typescript": "^6.0.3",
+    "vitest": "^5.0.1",
     "yargs": "^18.0.0",
-    "zod-validation-error": "^5.0.0",
-    "@langwatch/redaction": "0.0.1",
-    "@langwatch/ssrf": "0.0.1",
-    "@langwatch/handled-error": "0.1.0"
-  },
-  "dependencies": {
-    "@modelcontextprotocol/sdk": "^1.29.0",
-    "zod": "^4.3.6"
+    "zod-validation-error": "^5.0.0"
   },
-  "bin": {
-    "langwatch-mcp-server": "./dist/index.js"
+  "engines": {
+    "node": "^22.23.4 || ^24.22.1 || >=26.12.1"
   },
   "scripts": {
-    "start": "tsx src/index.ts",
+    "start": "node --experimental-transform-types src/index.ts",
     "build": "tsup && node build.js",
+    "typecheck": "GOMAXPROCS=2 GOMEMLIMIT=2GiB pnpm -w exec tsc -b mcp/typescript",
     "prepublish": "pnpm run build",
-    "test": "vitest"
+    "test": "vitest run",
+    "test:integration": "vitest run --config vitest.integration.config.js",
+    "test:scenario": "vitest run --config vitest.scenario.config.js"
   }
 }

package/pyproject.toml

@@ -5,12 +5,12 @@ description = "Add your description here"
 readme = "README.md"
 requires-python = ">=3.11"
 dependencies = [
-    "ipykernel>=6.30.1",
-    "langwatch",
-    "litellm>=1.84.0",
-    "mcp[cli]>=1.13.1",
-    "pandas>=2.3.2",
-    "tenacity>=9.1.2",
+  "ipykernel>=7.3.0",
+  "langwatch",
+  "litellm>=1.96.0",
+  "mcp[cli]>=2.0.0",
+  "pandas>=2.3.2",
+  "tenacity>=9.1.4",
 ]
 
 [tool.uv]
@@ -20,8 +20,8 @@ exclude-newer = "7 days"
 # package isn't a direct dependency, so a future re-resolution can't silently
 # pull a vulnerable version back in.
 constraint-dependencies = [
-    "python-dotenv>=1.2.2",  # security: python-dotenv advisory (#934)
-    "python-liquid>=2.2.1",  # security: python-liquid advisory (#1528)
+  "python-dotenv>=1.2.2", # security: python-dotenv advisory (#934)
+  "python-liquid>=2.2.1", # security: python-liquid advisory (#1528)
 ]
 
 [tool.uv.sources]

package/src/config.d.ts

@@ -1,13 +1,17 @@
 export interface McpConfig {
   apiKey: string | undefined;
   endpoint: string;
+  projectId?: string;
 }
 
 export declare function initConfig(args: {
   apiKey?: string;
   endpoint?: string;
+  projectId?: string;
 }): void;
 
+export declare function tryGetConfig(): McpConfig | undefined;
+
 export declare function getConfig(): McpConfig;
 
 export declare function requireApiKey(): string;

package/src/config.ts

@@ -6,19 +6,7 @@ export interface McpConfig {
   projectId?: string;
 }
 
-// ---------------------------------------------------------------------------
-// Singleton storage on globalThis
-//
-// When the langwatch app (CJS, no "type": "module") imports this module,
-// tsx creates a CJS-cached copy. When mcp-server's own ESM dist chunks
-// do `await import("./search-traces-*.js")`, Node creates a separate
-// ESM-cached copy. Module-level variables are NOT shared between them.
-//
-// By storing config on globalThis, both CJS and ESM instances read/write
-// the same object, fixing the "Config not initialized" error that occurs
-// when initConfig() runs on the CJS side but tool handlers execute on
-// the ESM side.
-// ---------------------------------------------------------------------------
+// Store on globalThis to share config between CJS and ESM module instances.
 
 const GLOBAL_KEY = "__langwatch_mcp_config" as const;
 const STORAGE_KEY = "__langwatch_mcp_config_storage" as const;
@@ -49,13 +37,10 @@ function getGlobalState(): McpGlobalState {
   };
 }
 
+/** Trim surrounding whitespace and drop any trailing slashes. */
 /**
- * Trim surrounding whitespace and drop any trailing slashes.
- *
- * Request URLs are built as `${endpoint}/api/...`, so an endpoint written as
- * `https://app.langwatch.ai/` would produce a double slash the router does not
- * match, and the caller gets an opaque 404 with nothing pointing at the
- * endpoint as the cause.
+ * Request URLs are built as `${endpoint}/api/...`; a trailing slash would
+ * double up and 404 with nothing pointing at the endpoint as the cause.
  */
 function normalizeEndpoint(endpoint: string): string {
   const trimmed = endpoint.trim();
@@ -64,7 +49,7 @@ function normalizeEndpoint(endpoint: string): string {
   return trimmed.slice(0, end);
 }
 
-export function initConfig(args: { apiKey?: string; endpoint?: string }): void {
+export function initConfig(args: { apiKey?: string; endpoint?: string; projectId?: string }): void {
   const state = getGlobalState();
   state.globalConfig = {
     apiKey: args.apiKey || process.env.LANGWATCH_API_KEY,
@@ -72,19 +57,25 @@ export function initConfig(args: { apiKey?: string; endpoint?: string }): void {
       normalizeEndpoint(args.endpoint ?? "") ||
       normalizeEndpoint(process.env.LANGWATCH_ENDPOINT ?? "") ||
       "https://app.langwatch.ai",
-    projectId: process.env.LANGWATCH_PROJECT_ID,
+    projectId: args.projectId ?? process.env.LANGWATCH_PROJECT_ID,
   };
 }
 
+// Current config: per-request scoped if inside runWithConfig(), otherwise global.
+// Returns undefined rather than throwing so callers don't need try/catch.
+export function tryGetConfig(): McpConfig | undefined {
+  const state = getGlobalState();
+  return state.configStorage.getStore() ?? state.globalConfig;
+}
+
+/** The current config: per-request scoped inside runWithConfig(), else global. */
 /**
- * Returns the current config: the per-request scoped config if inside
- * a `runWithConfig()` callback, otherwise the global config.
+ * Throws when there is none — every caller here needs it to proceed. A
+ * caller that can carry on without it asks `tryGetConfig()` instead.
  */
 export function getConfig(): McpConfig {
-  const state = getGlobalState();
-  const scoped = state.configStorage.getStore();
-  if (scoped) return scoped;
-  if (!state.globalConfig) {
+  const config = tryGetConfig();
+  if (!config) {
     console.error(
       "[MCP config] getConfig() failed: globalConfig is null, no scoped config active. " +
         "Was initConfig() called? Stack:",
@@ -92,7 +83,7 @@ export function getConfig(): McpConfig {
     );
     throw new Error("Config not initialized");
   }
-  return state.globalConfig;
+  return config;
 }
 
 export function requireApiKey(): string {
@@ -107,7 +98,7 @@ export function requireApiKey(): string {
       new Error().stack,
     );
     throw new Error(
-      "LANGWATCH_API_KEY is required. Set it via --apiKey flag or LANGWATCH_API_KEY environment variable."
+      "LANGWATCH_API_KEY is required. Set it via --apiKey flag or LANGWATCH_API_KEY environment variable.",
     );
   }
   return config.apiKey;