@iris-eval/mcp-server 0.1.2 → 0.1.4

package/dist/dashboard/assets/{index-DsCtYyvh.css → index-zuWgVxsu.css}

@@ -1 +1 @@
-:root{--bg-primary: #0a0a0b;--bg-secondary: #141416;--bg-tertiary: #1c1c1f;--bg-hover: #252528;--text-primary: #fafafa;--text-secondary: #a1a1aa;--text-muted: #71717a;--accent-primary: #6366f1;--accent-primary-hover: #818cf8;--accent-success: #22c55e;--accent-error: #ef4444;--accent-warning: #f59e0b;--accent-tool: #3b82f6;--accent-llm: #a855f7;--border-color: #27272a;--border-radius: 8px;--border-radius-sm: 4px;--border-radius-lg: 12px;--font-sans: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, sans-serif;--font-mono: "JetBrains Mono", "Fira Code", monospace;--font-size-xs: .75rem;--font-size-sm: .875rem;--font-size-base: 1rem;--font-size-lg: 1.125rem;--font-size-xl: 1.25rem;--font-size-2xl: 1.5rem;--font-size-3xl: 2rem;--space-1: .25rem;--space-2: .5rem;--space-3: .75rem;--space-4: 1rem;--space-5: 1.25rem;--space-6: 1.5rem;--space-8: 2rem;--space-10: 2.5rem;--space-12: 3rem;--shadow-sm: 0 1px 2px rgba(0, 0, 0, .3);--shadow-md: 0 4px 6px rgba(0, 0, 0, .4);--shadow-lg: 0 10px 15px rgba(0, 0, 0, .5);--transition-fast: .15s ease;--transition-base: .2s ease}*,*:before,*:after{box-sizing:border-box;margin:0;padding:0}html,body,#root{height:100%;width:100%}body{font-family:var(--font-sans);font-size:var(--font-size-base);color:var(--text-primary);background-color:var(--bg-primary);line-height:1.5;-webkit-font-smoothing:antialiased}a{color:var(--accent-primary);text-decoration:none}a:hover{color:var(--accent-primary-hover)}button{cursor:pointer;font-family:inherit}input,select{font-family:inherit;font-size:inherit}code,pre{font-family:var(--font-mono)}::-webkit-scrollbar{width:8px;height:8px}::-webkit-scrollbar-track{background:var(--bg-primary)}::-webkit-scrollbar-thumb{background:var(--border-color);border-radius:4px}::-webkit-scrollbar-thumb:hover{background:var(--text-muted)}
+:root{--bg-primary: #0a0a0b;--bg-secondary: #141416;--bg-tertiary: #1c1c1f;--bg-hover: #252528;--text-primary: #fafafa;--text-secondary: #a1a1aa;--text-muted: #71717a;--accent-primary: #0d9488;--accent-primary-hover: #14b8a6;--accent-success: #22c55e;--accent-error: #ef4444;--accent-warning: #f59e0b;--accent-tool: #3b82f6;--accent-llm: #a855f7;--border-color: #27272a;--border-radius: 8px;--border-radius-sm: 4px;--border-radius-lg: 12px;--font-sans: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, sans-serif;--font-mono: "JetBrains Mono", "Fira Code", monospace;--font-size-xs: .75rem;--font-size-sm: .875rem;--font-size-base: 1rem;--font-size-lg: 1.125rem;--font-size-xl: 1.25rem;--font-size-2xl: 1.5rem;--font-size-3xl: 2rem;--space-1: .25rem;--space-2: .5rem;--space-3: .75rem;--space-4: 1rem;--space-5: 1.25rem;--space-6: 1.5rem;--space-8: 2rem;--space-10: 2.5rem;--space-12: 3rem;--shadow-sm: 0 1px 2px rgba(0, 0, 0, .3);--shadow-md: 0 4px 6px rgba(0, 0, 0, .4);--shadow-lg: 0 10px 15px rgba(0, 0, 0, .5);--transition-fast: .15s ease;--transition-base: .2s ease}*,*:before,*:after{box-sizing:border-box;margin:0;padding:0}html,body,#root{height:100%;width:100%}body{font-family:var(--font-sans);font-size:var(--font-size-base);color:var(--text-primary);background-color:var(--bg-primary);line-height:1.5;-webkit-font-smoothing:antialiased}a{color:var(--accent-primary);text-decoration:none}a:hover{color:var(--accent-primary-hover)}button{cursor:pointer;font-family:inherit}input,select{font-family:inherit;font-size:inherit}code,pre{font-family:var(--font-mono)}::-webkit-scrollbar{width:8px;height:8px}::-webkit-scrollbar-track{background:var(--bg-primary)}::-webkit-scrollbar-thumb{background:var(--border-color);border-radius:4px}::-webkit-scrollbar-thumb:hover{background:var(--text-muted)}

package/dist/dashboard/favicon.svg

@@ -1,5 +1,6 @@
 <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 32 32" fill="none">
-  <circle cx="16" cy="16" r="14" stroke="#6366f1" stroke-width="2.5"/>
-  <circle cx="16" cy="16" r="7" fill="#6366f1" opacity="0.3"/>
-  <circle cx="16" cy="16" r="3.5" fill="#6366f1"/>
+  <circle cx="16" cy="16" r="14" stroke="#14b8a6" stroke-width="2.5"/>
+  <circle cx="16" cy="16" r="8" fill="#0d9488"/>
+  <circle cx="16" cy="16" r="3.5" fill="#0a0f0e"/>
+  <circle cx="13" cy="13.5" r="1.5" fill="white" opacity="0.6"/>
 </svg>

package/dist/dashboard/index.html

@@ -5,8 +5,8 @@
     <meta name="viewport" content="width=device-width, initial-scale=1.0" />
     <link rel="icon" type="image/svg+xml" href="/favicon.svg" />
     <title>Iris — Agent Eval & Observability</title>
-    <script type="module" crossorigin src="/assets/index-BgsGBkem.js"></script>
-    <link rel="stylesheet" crossorigin href="/assets/index-DsCtYyvh.css">
+    <script type="module" crossorigin src="/assets/index-CGWlxM5E.js"></script>
+    <link rel="stylesheet" crossorigin href="/assets/index-zuWgVxsu.css">
   </head>
   <body>
     <div id="root"></div>

package/dist/storage/sqlite-adapter.d.ts

@@ -1,4 +1,4 @@
-import type { IStorageAdapter, DashboardSummary, TraceQueryOptions, TraceQueryResult } from '../types/query.js';
+import type { IStorageAdapter, DashboardSummary, TraceQueryOptions, TraceQueryResult, EvalStatsPeriod, EvalStats, EvalStatsTrendBucket, EvalStatsRuleBreakdown, EvalStatsFailure } from '../types/query.js';
 import type { Trace, Span } from '../types/trace.js';
 import type { EvalResult } from '../types/eval.js';
 export declare class SqliteAdapter implements IStorageAdapter {
@@ -25,6 +25,11 @@ export declare class SqliteAdapter implements IStorageAdapter {
         total: number;
     }>;
     getDashboardSummary(sinceHours?: number): Promise<DashboardSummary>;
+    private periodToSince;
+    getEvalStats(period: EvalStatsPeriod): Promise<EvalStats>;
+    getEvalStatsTrend(period: EvalStatsPeriod): Promise<EvalStatsTrendBucket[]>;
+    getEvalStatsRules(period: EvalStatsPeriod): Promise<EvalStatsRuleBreakdown[]>;
+    getEvalStatsFailures(period: EvalStatsPeriod, limit: number): Promise<EvalStatsFailure[]>;
     deleteTracesOlderThan(days: number): Promise<number>;
     getDistinctValues(column: string): Promise<string[]>;
     private rowToTrace;

package/dist/storage/sqlite-adapter.js

@@ -167,6 +167,170 @@ export class SqliteAdapter {
             top_agents: topAgents,
         };
     }
+    // ---------------------------------------------------------------------------
+    // Eval-stats endpoints (v0.2.0 dashboard)
+    // ---------------------------------------------------------------------------
+    periodToSince(period) {
+        const hours = period === '24h' ? 24 : period === '7d' ? 168 : 720;
+        return new Date(Date.now() - hours * 60 * 60 * 1000).toISOString();
+    }
+    async getEvalStats(period) {
+        const since = this.periodToSince(period);
+        const agg = this.db.prepare(`
+      SELECT
+        COUNT(*)                                     AS total_evals,
+        COALESCE(AVG(score), 0)                      AS avg_score,
+        SUM(CASE WHEN passed = 1 THEN 1 ELSE 0 END) AS passed_count
+      FROM eval_results
+      WHERE created_at >= ?
+    `).get(since);
+        const cost = this.db.prepare(`
+      SELECT COALESCE(SUM(cost_usd), 0) AS total_cost
+      FROM traces
+      WHERE timestamp >= ?
+    `).get(since);
+        const agents = this.db.prepare(`
+      SELECT COUNT(DISTINCT agent_name) AS agent_count
+      FROM traces
+      WHERE timestamp >= ?
+    `).get(since);
+        // Safety violations — scan rule_results JSON for failing safety rules.
+        // rule_results is stored as a JSON array of EvalRuleResult objects.
+        const safetyRows = this.db.prepare(`
+      SELECT rule_results
+      FROM eval_results
+      WHERE created_at >= ?
+        AND eval_type = 'safety'
+        AND passed = 0
+    `).all(since);
+        const violations = { pii: 0, injection: 0, hallucination: 0 };
+        for (const row of safetyRows) {
+            const rules = JSON.parse(row.rule_results);
+            for (const r of rules) {
+                if (r.passed)
+                    continue;
+                if (r.ruleName === 'no_pii')
+                    violations.pii++;
+                if (r.ruleName === 'no_injection_patterns')
+                    violations.injection++;
+                if (r.ruleName === 'no_hallucination_markers')
+                    violations.hallucination++;
+            }
+        }
+        return {
+            passRate: agg.total_evals > 0
+                ? Math.round((agg.passed_count / agg.total_evals) * 1000) / 1000
+                : 0,
+            avgScore: Math.round(agg.avg_score * 1000) / 1000,
+            totalEvals: agg.total_evals,
+            safetyViolations: violations,
+            totalCost: Math.round(cost.total_cost * 100) / 100,
+            agentCount: agents.agent_count,
+            period,
+        };
+    }
+    async getEvalStatsTrend(period) {
+        const since = this.periodToSince(period);
+        // Determine bucket format for strftime
+        let bucketExpr;
+        if (period === '24h') {
+            // hourly buckets
+            bucketExpr = "strftime('%Y-%m-%dT%H:00:00Z', created_at)";
+        }
+        else if (period === '7d') {
+            // 6-hour buckets: floor hour to nearest 6
+            bucketExpr =
+                "strftime('%Y-%m-%dT', created_at) || printf('%02d', (CAST(strftime('%H', created_at) AS INTEGER) / 6) * 6) || ':00:00Z'";
+        }
+        else {
+            // daily buckets
+            bucketExpr = "strftime('%Y-%m-%dT00:00:00Z', created_at)";
+        }
+        const rows = this.db.prepare(`
+      SELECT
+        ${bucketExpr}                                  AS bucket,
+        COALESCE(AVG(score), 0)                        AS avg_score,
+        CASE WHEN COUNT(*) > 0
+          THEN CAST(SUM(CASE WHEN passed = 1 THEN 1 ELSE 0 END) AS REAL) / COUNT(*)
+          ELSE 0 END                                   AS pass_rate,
+        COUNT(*)                                       AS eval_count
+      FROM eval_results
+      WHERE created_at >= ?
+      GROUP BY bucket
+      ORDER BY bucket
+    `).all(since);
+        return rows.map((r) => ({
+            timestamp: r.bucket,
+            avgScore: Math.round(r.avg_score * 1000) / 1000,
+            passRate: Math.round(r.pass_rate * 1000) / 1000,
+            evalCount: r.eval_count,
+        }));
+    }
+    async getEvalStatsRules(period) {
+        const since = this.periodToSince(period);
+        const rows = this.db.prepare(`
+      SELECT rule_results
+      FROM eval_results
+      WHERE created_at >= ?
+    `).all(since);
+        // Aggregate per-rule stats from the JSON arrays
+        const ruleMap = new Map();
+        for (const row of rows) {
+            const rules = JSON.parse(row.rule_results);
+            for (const r of rules) {
+                const entry = ruleMap.get(r.ruleName) ?? { totalRun: 0, failCount: 0 };
+                entry.totalRun++;
+                if (!r.passed)
+                    entry.failCount++;
+                ruleMap.set(r.ruleName, entry);
+            }
+        }
+        const result = [];
+        for (const [rule, stats] of ruleMap) {
+            result.push({
+                rule,
+                passRate: stats.totalRun > 0
+                    ? Math.round(((stats.totalRun - stats.failCount) / stats.totalRun) * 1000) / 1000
+                    : 0,
+                totalRun: stats.totalRun,
+                failCount: stats.failCount,
+            });
+        }
+        // Sort by passRate ASC (worst rules first)
+        result.sort((a, b) => a.passRate - b.passRate);
+        return result;
+    }
+    async getEvalStatsFailures(period, limit) {
+        const since = this.periodToSince(period);
+        const rows = this.db.prepare(`
+      SELECT
+        e.trace_id,
+        COALESCE(t.agent_name, 'unknown') AS agent_name,
+        e.rule_results,
+        e.score,
+        e.output_text,
+        e.created_at
+      FROM eval_results e
+      LEFT JOIN traces t ON t.trace_id = e.trace_id
+      WHERE e.created_at >= ?
+        AND e.passed = 0
+      ORDER BY e.created_at DESC
+      LIMIT ?
+    `).all(since, limit);
+        return rows.map((r) => {
+            // Find the first failing rule to surface as the primary rule
+            const rules = JSON.parse(r.rule_results);
+            const failingRule = rules.find((rule) => !rule.passed);
+            return {
+                traceId: r.trace_id ?? '',
+                agent: r.agent_name,
+                rule: failingRule?.ruleName ?? 'unknown',
+                score: Math.round(r.score * 1000) / 1000,
+                output: r.output_text.length > 200 ? r.output_text.slice(0, 200) + '...' : r.output_text,
+                timestamp: r.created_at,
+            };
+        });
+    }
     async deleteTracesOlderThan(days) {
         const cutoff = new Date(Date.now() - days * 24 * 60 * 60 * 1000).toISOString();
         const result = this.db.prepare('DELETE FROM traces WHERE timestamp < ?').run(cutoff);

package/dist/transport/http.js

@@ -6,8 +6,15 @@ import { createErrorHandler } from '../middleware/error-handler.js';
 import { createMcpRateLimiter } from '../middleware/rate-limit.js';
 export async function createHttpTransport(mcpServer, config, logger) {
     const app = express();
-    // Security headers (no CSP — API only, no HTML)
-    app.use(helmet({ contentSecurityPolicy: false }));
+    // Security headers — API-only server, restrictive CSP
+    app.use(helmet({
+        contentSecurityPolicy: {
+            directives: {
+                defaultSrc: ["'none'"],
+                frameAncestors: ["'none'"],
+            },
+        },
+    }));
     // Body parser with size limit
     app.use(express.json({ limit: config.security.requestSizeLimit }));
     // Health endpoint (no auth, no rate limit)

package/dist/types/index.d.ts

@@ -1,4 +1,4 @@
 export type { SpanKind, SpanStatus, SpanEvent, ToolCallRecord, TokenUsage, Span, Trace, } from './trace.js';
 export type { EvalType, EvalRule, EvalContext, EvalRuleResult, EvalResult, CustomRuleType, CustomRuleDefinition, } from './eval.js';
-export type { TraceFilter, TraceQueryOptions, TraceQueryResult, DashboardSummary, IStorageAdapter, } from './query.js';
+export type { TraceFilter, TraceQueryOptions, TraceQueryResult, DashboardSummary, EvalStatsPeriod, EvalStats, EvalStatsTrendBucket, EvalStatsRuleBreakdown, EvalStatsFailure, IStorageAdapter, } from './query.js';
 export type { IrisConfig } from './config.js';

package/dist/types/query.d.ts

@@ -22,6 +22,40 @@ export interface TraceQueryResult {
     limit: number;
     offset: number;
 }
+export type EvalStatsPeriod = '24h' | '7d' | '30d';
+export interface EvalStats {
+    passRate: number;
+    avgScore: number;
+    totalEvals: number;
+    safetyViolations: {
+        pii: number;
+        injection: number;
+        hallucination: number;
+    };
+    totalCost: number;
+    agentCount: number;
+    period: EvalStatsPeriod;
+}
+export interface EvalStatsTrendBucket {
+    timestamp: string;
+    avgScore: number;
+    passRate: number;
+    evalCount: number;
+}
+export interface EvalStatsRuleBreakdown {
+    rule: string;
+    passRate: number;
+    totalRun: number;
+    failCount: number;
+}
+export interface EvalStatsFailure {
+    traceId: string;
+    agent: string;
+    rule: string;
+    score: number;
+    output: string;
+    timestamp: string;
+}
 export interface DashboardSummary {
     total_traces: number;
     avg_latency_ms: number;
@@ -61,4 +95,8 @@ export interface IStorageAdapter {
     getDashboardSummary(sinceHours?: number): Promise<DashboardSummary>;
     deleteTracesOlderThan(days: number): Promise<number>;
     getDistinctValues(column: string): Promise<string[]>;
+    getEvalStats(period: EvalStatsPeriod): Promise<EvalStats>;
+    getEvalStatsTrend(period: EvalStatsPeriod): Promise<EvalStatsTrendBucket[]>;
+    getEvalStatsRules(period: EvalStatsPeriod): Promise<EvalStatsRuleBreakdown[]>;
+    getEvalStatsFailures(period: EvalStatsPeriod, limit: number): Promise<EvalStatsFailure[]>;
 }

package/package.json

@@ -1,7 +1,8 @@
 {
   "name": "@iris-eval/mcp-server",
-  "version": "0.1.2",
-  "description": "MCP-native agent evaluation and observability server",
+  "version": "0.1.4",
+  "description": "The agent eval standard for MCP — score output quality, catch safety failures, enforce cost budgets",
+  "mcpName": "io.github.iris-eval/mcp-server",
   "type": "module",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
@@ -20,17 +21,31 @@
     "test:watch": "vitest",
     "test:coverage": "vitest run --coverage",
     "test:integration": "vitest run tests/integration/",
+    "version:check": "bash scripts/check-version.sh",
+    "version:sync": "node scripts/sync-versions.mjs",
     "clean": "rm -rf dist coverage",
     "seed:demo": "tsx scripts/seed-demo-data.ts",
-    "demo": "tsx scripts/demo.ts"
+    "demo": "tsx scripts/demo.ts",
+    "postinstall": "echo \"\\n  ✅ Iris installed — the agent eval standard for MCP\\n  📖 Docs: https://iris-eval.com\\n  🎯 Try the playground: https://iris-eval.com/playground\\n  ⭐ Star us: https://github.com/iris-eval/mcp-server\\n\""
   },
   "keywords": [
     "mcp",
+    "mcp-server",
+    "model-context-protocol",
     "agent",
+    "ai-agent",
     "evaluation",
     "observability",
     "tracing",
-    "llm"
+    "llm",
+    "claude",
+    "agent-evaluation",
+    "ai-observability",
+    "pii-detection",
+    "cost-tracking",
+    "agent-eval",
+    "output-quality",
+    "quality-gate"
   ],
   "author": "",
   "license": "MIT",
@@ -38,7 +53,7 @@
     "type": "git",
     "url": "git+https://github.com/iris-eval/mcp-server.git"
   },
-  "homepage": "https://github.com/iris-eval/mcp-server#readme",
+  "homepage": "https://iris-eval.com",
   "bugs": {
     "url": "https://github.com/iris-eval/mcp-server/issues"
   },
@@ -53,7 +68,7 @@
   },
   "dependencies": {
     "@modelcontextprotocol/sdk": "^1.27.0",
-    "better-sqlite3": "^11.0.0",
+    "better-sqlite3": "^12.8.0",
     "express": "^5.1.0",
     "express-rate-limit": "^8.3.1",
     "helmet": "^8.1.0",
@@ -65,7 +80,7 @@
     "@types/better-sqlite3": "^7.6.0",
     "@types/express": "^5.0.0",
     "@types/node": "^22.0.0",
-    "@typescript-eslint/eslint-plugin": "^8.57.0",
+    "@typescript-eslint/eslint-plugin": "^8.57.1",
     "@typescript-eslint/parser": "^8.57.0",
     "@vitest/coverage-v8": "^3.0.0",
     "eslint": "^9.0.0",

package/server.json

@@ -1,69 +1,43 @@
 {
-  "name": "iris-eval",
-  "version": "0.1.0",
-  "description": "MCP-native agent evaluation and observability server",
-  "homepage": "https://github.com/iris-eval/mcp-server",
-  "packages": {
-    "npm": {
-      "name": "@iris-eval/mcp-server",
-      "registry": "https://registry.npmjs.org"
-    },
-    "oci": {
-      "name": "ghcr.io/iris-eval/mcp-server",
-      "registry": "ghcr.io"
-    }
+  "$schema": "https://static.modelcontextprotocol.io/schemas/2025-12-11/server.schema.json",
+  "name": "io.github.iris-eval/mcp-server",
+  "description": "MCP-native agent evaluation and observability server for AI agents",
+  "repository": {
+    "url": "https://github.com/iris-eval/mcp-server",
+    "source": "github"
   },
-  "tools": [
-    {
-      "name": "log_trace",
-      "description": "Log an agent execution trace with spans, tool calls, and metrics"
-    },
+  "version": "0.1.4",
+  "packages": [
     {
-      "name": "evaluate_output",
-      "description": "Evaluate agent output quality using configurable rules"
-    },
-    {
-      "name": "get_traces",
-      "description": "Query stored traces with filters, pagination, and summary stats"
-    }
-  ],
-  "resources": [
-    {
-      "uri": "iris://dashboard/summary",
-      "description": "Dashboard summary with key metrics and trends"
-    },
-    {
-      "uri_template": "iris://traces/{trace_id}",
-      "description": "Full trace detail with spans and evaluation results"
-    }
-  ],
-  "env": {
-    "IRIS_TRANSPORT": {
-      "description": "Transport type: stdio or http",
-      "default": "stdio"
-    },
-    "IRIS_PORT": {
-      "description": "HTTP transport port",
-      "default": "3000"
-    },
-    "IRIS_DB_PATH": {
-      "description": "SQLite database path",
-      "default": "~/.iris/iris.db"
-    },
-    "IRIS_LOG_LEVEL": {
-      "description": "Log level: debug, info, warn, error",
-      "default": "info"
-    },
-    "IRIS_DASHBOARD": {
-      "description": "Enable web dashboard",
-      "default": "false"
-    },
-    "IRIS_API_KEY": {
-      "description": "API key for HTTP authentication (optional, recommended for production)"
-    },
-    "IRIS_ALLOWED_ORIGINS": {
-      "description": "Comma-separated list of allowed CORS origins",
-      "default": "http://localhost:*"
+      "registryType": "npm",
+      "identifier": "@iris-eval/mcp-server",
+      "version": "0.1.3",
+      "transport": {
+        "type": "stdio"
+      },
+      "environmentVariables": [
+        {
+          "description": "API key for HTTP authentication (optional, recommended for production)",
+          "isRequired": false,
+          "format": "string",
+          "isSecret": true,
+          "name": "IRIS_API_KEY"
+        },
+        {
+          "description": "SQLite database path",
+          "isRequired": false,
+          "format": "string",
+          "isSecret": false,
+          "name": "IRIS_DB_PATH"
+        },
+        {
+          "description": "Log level: debug, info, warn, error",
+          "isRequired": false,
+          "format": "string",
+          "isSecret": false,
+          "name": "IRIS_LOG_LEVEL"
+        }
+      ]
     }
-  }
+  ]
 }