MCP Foundry

Turn any codebase into a working MCP (Model Context Protocol) server. Paste a GitHub repo URL or upload code → Foundry discovers the capabilities → you curate which to expose (risk-flagged, destructive opt-in) → it generates a runnable FastMCP server + connect config.

This is the web platform built on top of the codebase-to-mcp skill's engine logic.

Deep-understanding report

Before (or instead of) generating a server, Foundry can read the repo in depth and explain it: after analyzing, click "Generate deep report" for a narrative breakdown — overview, architecture, how it works, key modules, and where to start building on top.

• With ANTHROPIC_API_KEY set, Claude writes the narrative, grounded only in the repo's files/README/capabilities (no fabrication).
• Without a key, it falls back to a structural report built straight from the discovery data.

Set the key (optional) via a local .env (see .env.example) — it's server-side, never shipped to the browser.

Run it

python3 -m venv .venv
.venv/bin/pip install -r requirements.txt
.venv/bin/python -m uvicorn app.main:app --reload --port 8099
# open http://127.0.0.1:8099

Layout

mcp-foundry/
├── engine/
│   ├── discovery.py    # scan a repo -> capabilities (FastAPI/Flask/Django/plain-python, AST-based)
│   ├── generator.py    # capabilities -> FastMCP server bundle (http-proxy or in-process)
│   └── analyzer.py     # repo + discovery -> deep-understanding report (Claude or structural)
├── app/
│   ├── main.py         # FastAPI: /api/analyze, /api/report, /api/generate, /api/download
│   └── templates/index.html   # single-page UI (input -> curate -> generate)
├── tests/
│   ├── test_discovery.py / test_generator.py   # engine unit checks
│   └── e2e_test.py     # end-to-end against the running server
├── CONTRACT.md         # the shared spec (Capability shape + API)
└── ROADMAP.md          # v1 scope + phases 2-4 (auth, hosting, more inputs)

Status

v1: the core loop works end-to-end (verified — upload/clone → analyze → curate → generate → download, with the emitted server.py compiling). Single-tenant, in-memory, generates code you run. Auth, persistence, and managed hosting are Phase 2–3 — see ROADMAP.md.

Test

.venv/bin/python tests/test_discovery.py
.venv/bin/python tests/test_generator.py
.venv/bin/python tests/e2e_test.py     # requires the server running on :8099