@tracepact/mcp-server

MCP (Model Context Protocol) server that exposes TracePact as tools for agentic IDEs.

Installation

npm install -g @tracepact/mcp-server

Usage

The server communicates via stdio and is designed to be registered in your IDE's MCP configuration.

Claude Desktop / Claude Code

Add to your MCP config:

{
  "mcpServers": {
    "tracepact": {
      "command": "tracepact-mcp-server"
    }
  }
}

The server provides instructions that guide agents on the recommended tool usage order.

Available Tools

tracepact_audit

Static analysis of a SKILL.md file. No API key needed.

Input:

• skill_path — Path to the SKILL.md file

Output:

• riskLevel — "none" | "low" | "medium" | "high" | "critical"
• pass — Whether the audit passed
• findings — Array of issues found
• summary — Counts by severity (critical, high, medium, low)

Checks for risky tool combinations, prompt hygiene, skill completeness, and opaque tools.

tracepact_run

Execute TracePact tests via Vitest.

Input:

• skill_path — Path to SKILL.md or test directory
• live (optional) — Run against real LLM APIs (default: false)
• provider (optional) — LLM provider name
• budget (optional) — Max token budget

Output:

• pass — Whether all tests passed
• output — Vitest JSON output
• error — Error message if failed

Timeout: 120 seconds.

tracepact_capture

Auto-generate a test file from a recorded cassette.

Input:

• skill_path — Path to SKILL.md
• prompt — Representative prompt for assertion inference

Output:

• testFile — Generated test code
• cassettePath — Path to cassette used
• assertionsGenerated — Number of assertions inferred

tracepact_replay

Replay a cassette without calling any API.

Input:

• cassette_path — Path to cassette JSON file

Output:

• pass — Whether replay succeeded
• trace — Full trace object with calls, totalCalls, totalDurationMs

tracepact_diff

Compare two cassette recordings to detect behavioral drift.

Input:

• cassette_a — Baseline cassette path (before change)
• cassette_b — Comparison cassette path (after change)

Output:

• changed — Whether behavior differs
• additions — Tool calls in B but not A
• removals — Tool calls in A but not B
• diffs — Argument changes per tool call

tracepact_list_tests

Find test files and cassettes associated with a skill.

Input:

• skill_path — Path to SKILL.md

Output:

• tests — Array of {path, name} for .test.ts / .test.js files
• cassettes — Array of cassette files found in cassettes/ or __cassettes__/

Recommended Workflow

A typical agentic IDE workflow:

1. tracepact_audit — Analyze the skill file for issues
2. tracepact_list_tests — Discover existing tests and cassettes
3. tracepact_run — Execute the test suite
4. tracepact_capture — Generate new tests from cassettes
5. tracepact_diff — Compare cassettes after changes
6. tracepact_replay — Quick validation without API calls

License

MIT