Skip to main content
Glama

Trace — the CV tool that won't lie for you

An MCP server that turns your own Claude (Desktop, Code, or any MCP client) into an honest CV co-pilot. Your career data stays on your machine, and every claim on a generated CV is mechanically verified against what you actually did.

Why this exists

Every AI CV tool has the same failure mode: it invents. Trace inverts it.

  1. Truth Base — your CV is parsed into a structured record of what you actually did (~/.trace/truth_base.json, never uploaded anywhere).

  2. Enrich interview — sharp questions that surface the metrics and scope your CV under-sells. Facts you supply, never suggestions to inflate.

  3. Honest fit score — a job description is scored against your Truth Base with a verdict of apply, stretch, or skip. Trace will tell you not to apply.

  4. Tailor with receipts — the CV is generated ONLY from Truth Base evidence. Every claim carries a JSON-path receipt, and the server's validate_receipts tool resolves each path mechanically. A claim that doesn't resolve fails the gate; the model must remove it before you ever see the document.

The honesty guardrail is code, not a system prompt.

Related MCP server: tech-memory-mcp

Install

Requires Python 3.10+ on PATH (python.org/downloads; on Windows tick "Add python.exe to PATH"). No API key: the model you're already talking to does the generation.

Claude Desktop — automatic (one command)

Paste one line into a terminal and everything is done for you (download, install, config registration with a backup of your existing config):

Windows (PowerShell):

powershell -ExecutionPolicy Bypass -c "irm https://raw.githubusercontent.com/defibabylon/trace-mcp/main/install.ps1 | iex"

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/defibabylon/trace-mcp/main/install.sh | bash

Restart Claude Desktop, then say "run the Trace wizard".

Claude Desktop — one click (.mcpb)

  1. Download trace-mcp.mcpb from the latest release (dependencies are bundled inside).

  2. Double-click the file, or in Claude Desktop go to Settings → Extensions and drop it there.

  3. Click Install, then start a new chat and say "run the Trace wizard".

Claude Desktop (manual config)

Clone the repo, run pip install -r requirements.txt, then add to claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "trace": {
      "command": "python",
      "args": ["<absolute path>/trace-mcp/server.py"]
    }
  }
}

Restart Claude Desktop; the Trace tools appear under the 🔌 icon.

Claude Code

git clone https://github.com/defibabylon/trace-mcp.git
cd trace-mcp && pip install -r requirements.txt
claude mcp add trace -- python server.py

Smithery

From the Smithery listing or CLI: npx -y smithery@latest mcp add defibabylon/trace-mcp --client claude

Use

Start with the trace_wizard prompt (or just say "run the Trace wizard"). Steps:

Prompt

What happens

trace_wizard

Resume wherever you left off

parse_cv

Paste your CV → Truth Base

enrich

Interview to fill gaps (metrics, scope, your voice)

score_fit

Paste a JD → blockers, gap map, apply/stretch/skip

tailor

Receipt-verified CV + cover letter, exported to ~/.trace/jobs/<company>/

prep_pack

Interview prep: company cheat sheet, question bank, a real Anki deck (.apkg), and a morning-of briefing script

Say "I applied" / "they invited me to interview" and the built-in tracker (set_status) keeps every application's status; list_jobs shows the whole pipeline.

Obsidian career vault

Say "set up Obsidian" and Trace turns its folder into a proper Obsidian vault: a Home.md dashboard (every application with fit, status, and linked documents) and a readable Truth Base.md, both auto-refreshed on every change. Open the folder as a vault and your job hunt has persistent, browsable memory outside chat. Your own notes are never touched; Trace only rewrites those two files.

Tools (for the model)

get_wizard_state, save_truth_base, get_truth_base, save_job, get_job, record_fit, validate_receipts, export_document, list_jobs, set_status, build_anki, setup_obsidian

Privacy Policy

Trace collects nothing: no telemetry, no accounts, no network requests of its own. Everything lives in ~/.trace/ on your machine (override with TRACE_HOME); nothing is sent anywhere except the conversation you are already having with your model. Delete the folder to reset. Full policy: PRIVACY.md

Test

python test_server.py

Build + publish the MCPB bundle (Smithery)

pip install --target lib mcp genanki
python build_bundle.py
npx -y smithery@latest mcp publish ./trace-mcp.mcpb -n defibabylon/trace-mcp

The bundle vendors the mcp dependency into lib/ (see manifest.json), so users only need Python 3.10+ on PATH. We zip directly (build_bundle.py) rather than using mcpb pack: Smithery's registry requires MCP-spec tool inputSchema objects in the manifest, which mcpb pack's stricter schema rejects.

Live listing: https://smithery.ai/servers/defibabylon/trace-mcp

License

MIT — AtlasFlow Systems. Trace is an AtlasFlow product; the full web app (no MCP client needed) is in development.

Related MCP Connectors

Related MCP Servers

  • A
    license
    A
    quality
    A
    maintenance
    An MCP server that gives Claude persistent, locally-stored memory with verification-gated lessons and a Pattern Oracle to prevent unverified guesses from becoming ingrained facts.
    13
    73 npm
    1
    MIT
  • A
    license
    Not graded
    quality
    D
    maintenance
    A MCP server that gives Claude Code and other AI assistants long-term memory by automatically extracting technical knowledge from conversations and retrieving relevant experiences in future sessions.
    6 npm
    MIT
  • A
    license
    A
    quality
    A
    maintenance
    An MCP server that exposes a perpetual, honest job-application pipeline as typed tools an LLM agent can call, with fit scoring, verified resume building, and a submission planner enforced by code, not prompts.
    16
    MIT