About sourcemap-retrace-mcp
sourcemap-retrace-mcp is an MCP server for developers investigating production errors in minified JavaScript. It maps stack frames to original TypeScript locations, retrieves nearby source context, and checks sourcemap validity.
Key features
- 01Map minified stack frames to TypeScript locations
- 02Retrieve surrounding source lines with pointers
- 03Audit sourcemap presence and source paths
Use cases
- 01Trace a production error to its source location
- 02Inspect code surrounding a mapped failure
- 03Check sourcemaps before deployment
MCP support
Exposes stack retracing, source context retrieval, and sourcemap auditing tools to MCP clients.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
7Saved tool list · updated 3w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| stackTrace | string | Yes | The raw minified error stack trace (multi-line string) |
| sourcemapDir | string | Yes | Directory containing .map files (or the dist directory where .js.map files live) |
Details
From the project README🗺️ sourcemap-retrace-mcp
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 — using source maps. Before your agent spends 20 minutes guessing at obfuscated code.
🤔 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.
🛠️ 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 — with line numbers and a column pointer.
Code Context
File: src/auth/session.ts
Target: line 142, column 8
139 │ const session = await getSession(token);
140 │ if (!session) throw new AuthError("invalid token");
141 │
> 142 │ return session.userId;
│ ^
143 │ }
144 │
145 │ 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
✓ main.a3f2c1.js — 47 source(s) mapped
✓ vendor.d4e5f6.js — 312 source(s) mapped
✓ worker.b1c2d3.js — 8 source(s) mapped
⚠ legacy.e7f8a9.js — 2 source(s) not found on disk:
../src/utils/deprecated.ts
../src/utils/compat.ts
⚡ Setup
{
"mcpServers": {
"sourcemap-retrace": {
"command": "npx",
"args": ["-y", "sourcemap-retrace-mcp"]
}
}
}
🚀 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 — scan for secrets before deploying
- release-readiness-triage-mcp — CI health check before release
📦 Links
License
MIT
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.