AI Job Application Tracker Agent

An agent that reads a job posting, extracts requirements via MCP tools, compares them with the resume profile, produces a fit score with an explanation (go/no-go), and logs the decision to a persistent application tracker.

Full technical scope — architecture, MCP tool contracts, data schemas, day-by-day plan — is in SPEC.md.

Status

This is the project skeleton after day 1: repository structure, Pydantic data schemas, and an MCP server with registered but not yet implemented tools (each one calls NotImplementedError with a note on what needs to be written). The actual extraction/scoring logic, LangGraph graph, tracker DB, and eval set are next steps per the plan in SPEC.md.

Related MCP server: witness-mcp

Setup

python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env  # заполнить ANTHROPIC_API_KEY

Fill data/resume_profile.json with your real skills (it currently contains a template with placeholders) — the adequacy of the fit score depends on it.

Checking that the MCP server starts

python -m mcp_server.server

Structure

job-tracker-agent/
├── SPEC.md                 # архитектура, контракты tools, план
├── mcp_server/
│   ├── server.py           # регистрация MCP tools (FastMCP)
│   ├── schemas.py          # Pydantic-модели
│   └── tools/
│       ├── extraction.py   # fetch_job_posting, extract_job_requirements
│       ├── scoring.py      # load_resume_profile, compute_fit_score
│       └── tracker.py      # log_application, update_application_outcome, query_tracker_stats
├── agent/
│   ├── state.py            # LangGraph state
│   └── graph.py            # узлы графа (заготовка)
├── data/
│   └── resume_profile.json # заполнить своими данными
└── evals/
    ├── golden_cases.yaml   # golden-кейсы для eval (заготовка)
    └── run_eval.py