Skip to main content
Glama
vola-trebla

sourcemap-retrace-mcp

by vola-trebla
README.md
# πŸ—ΊοΈ sourcemap-retrace-mcp

[![npm](https://img.shields.io/npm/v/sourcemap-retrace-mcp)](https://www.npmjs.com/package/sourcemap-retrace-mcp)
[![CI](https://github.com/vola-trebla/sourcemap-retrace-mcp/actions/workflows/ci.yml/badge.svg)](https://github.com/vola-trebla/sourcemap-retrace-mcp/actions)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)

**Your app crashed in production. The stack trace is useless. Your AI agent has no idea where the bug is.**

MCP server that decodes minified production stack traces back to original TypeScript source files, lines, and columns β€” using source maps. Before your agent spends 20 minutes guessing at obfuscated code.

---

## πŸ€” The problem

Your monitoring tool sends you this:

```
TypeError: Cannot read properties of undefined (reading 'userId')
    at e.<anonymous> (main.a3f2c1.js:1:47821)
    at h (vendor.d4e5f6.js:1:12045)
    at processQueue (main.a3f2c1.js:1:89234)
```

You ask your agent to debug it. The agent reads `main.a3f2c1.js`. It's 40,000 characters of minified JavaScript. It has no idea what `e.<anonymous>` at column 47821 means.

`sourcemap-retrace-mcp` maps that back to:

```
    at validateSession (src/auth/session.ts:142:8)
    at handleRequest (src/api/middleware.ts:67:3)
```

Now the agent knows exactly where to look.

---

## πŸ› οΈ Tools

### `retrace_stack`

Decode a minified stack trace back to original TypeScript source locations. Pass the raw stack trace and the directory containing your `.js.map` files.

```
Retrace Results
  Frames mapped: 3  |  Unmapped: 1

TypeError: Cannot read properties of undefined (reading 'userId')
    at validateSession (src/auth/session.ts:142:8)
    at handleRequest (src/api/middleware.ts:67:3)
    at processQueue (src/queue/processor.ts:28:12)
    at Module._compile (node:internal/modules/cjs/loader:1376:14)
```

### `retrieve_code_context`

Show the original source lines surrounding a mapped error location β€” with line numbers and a column pointer.

```
Code Context
  File: src/auth/session.ts
  Target: line 142, column 8

  139 β”‚   const session = await getSession(token);
  140 β”‚   if (!session) throw new AuthError("invalid token");
  141 β”‚
> 142 β”‚   return session.userId;
        β”‚         ^
  143 β”‚ }
  144 β”‚
  145 β”‚ export async function refreshSession(token: string) {
```

### `audit_sourcemap_match`

Validate that `.map` files in your dist directory are present and point to source files that actually exist on disk. Catches stale or missing maps before deployment.

```
Sourcemap Audit
  Directory: dist/
  Files checked: 4
  OK: 3  |  Missing maps: 0  |  Broken sources: 1

  βœ“ main.a3f2c1.js β€” 47 source(s) mapped
  βœ“ vendor.d4e5f6.js β€” 312 source(s) mapped
  βœ“ worker.b1c2d3.js β€” 8 source(s) mapped
  ⚠ legacy.e7f8a9.js β€” 2 source(s) not found on disk:
      ../src/utils/deprecated.ts
      ../src/utils/compat.ts
```

---

## ⚑ Setup

```json
{
  "mcpServers": {
    "sourcemap-retrace": {
      "command": "npx",
      "args": ["-y", "sourcemap-retrace-mcp"]
    }
  }
}
```

---

## πŸš€ Usage

> "I have a production error. Here's the stack trace: [paste]. My dist files are in /path/to/dist. Retrace it, show me the code around the error, and tell me what's wrong."

The agent runs `retrace_stack`, then `retrieve_code_context` on the mapped location, and can finally read the actual TypeScript that crashed.

Works great alongside:

- [env-secret-exposure-analyzer-mcp](https://www.npmjs.com/package/env-secret-exposure-analyzer-mcp) β€” scan for secrets before deploying
- [release-readiness-triage-mcp](https://www.npmjs.com/package/release-readiness-triage-mcp) β€” CI health check before release

---

## πŸ“¦ Links

- **npm:** [npmjs.com/package/sourcemap-retrace-mcp](https://www.npmjs.com/package/sourcemap-retrace-mcp)
- **GitHub:** [github.com/vola-trebla/sourcemap-retrace-mcp](https://github.com/vola-trebla/sourcemap-retrace-mcp)

## License

MIT

TDQS

A3.7/5.0

Scored across 7 tools

Disambiguation4/5

Most tools have clearly distinct purposes, but retrace_stack and ingest_telemetry_payload both produce retraced frames, and audit_sourcemap_match and verify_debug_id_integrity both validate sourcemap integrity, which could cause some confusion. However, descriptions clarify the differences well.

Naming Consistency5/5

All tool names follow a consistent verb_noun pattern with underscores, e.g., retrace_stack, audit_sourcemap_match, surface_async_causality. The verbs are specific and the nouns clearly indicate the target, making the set predictable and readable.

Tool Count5/5

With 7 tools, the server is well-scoped for sourcemap retracing and analysis. Each tool addresses a meaningful aspect of the workflow without excessive overlap or unnecessary bloat.

Completeness4/5

The tool set covers core retracing, context retrieval, validation, path normalization, telemetry ingestion, and async tracing. Minor gaps such as direct support for sourcemap generation or batch processing are not essential for the stated purpose, so the surface is largely complete.

Maintenance

ActivityInactive
ResponsivenessUnresponsive