A Model Context Protocol server that exposes the RE-Library knowledge base as a set of tools for any MCP-compatible client (Claude Desktop, Cursor, Continue, …).

The server reads the same content/**/*.md files that the GitHub Pages site reads, parses them, and serves them over JSON-RPC via stdio. No hosted backend; runs on the analyst's box.

Install

pip install re-library-mcp

For development, install from the repo in editable mode:

git clone https://github.com/Heretek-AI/RE-Library
cd RE-Library/mcp-server
pip install -e ".[dev]"

Use

From Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "re-library": {
      "command": "python",
      "args": ["-m", "re_library_mcp"]
    }
  }
}

Restart Claude Desktop. You'll see re-library listed with five tools available. Try asking:

"What are the main ways to bypass SSL pinning on Android?"

From Cursor / Continue / any MCP client

Same pattern — point the client at python -m re_library_mcp over stdio. See your client's docs for the exact config key.

Tools

Tool	Purpose
`search_re(query, category?, platform?, max_results?)`	Free-text search; returns up to 5 hits with title, category, snippet, and score.
`get_entry(slug)`	Full text of one entry. The slug is `<category>/<NN>-<slug>`, e.g. `android/01-apk-structure`.
`list_categories()`	All eight categories with their entry counts.
`list_entries(category)`	Lightweight summary of every entry in one category.
`get_anti_analysis_techniques(platform?)`	Convenience aggregator for the `anti-analysis` category.

Configuration

Env var	Default	Notes
`RE_LIBRARY_CONTENT_DIR`	(unset)	Path to a local copy of `content/`. If set, the loader reads from disk — useful for development and air-gapped use.
`RE_LIBRARY_REPO`	`Heretek-AI/RE-Library`	GitHub `owner/name` to fetch from when no local content is set.
`RE_LIBRARY_BRANCH`	`main`	Branch to fetch from.
`RE_LIBRARY_OFFLINE`	(unset)	If truthy, never make network requests; fail if local content is missing.

Smoke test

Verify the install and content resolution without starting a JSON-RPC session:

re-library-mcp --check
# or
python -m re_library_mcp --check

This prints a one-line summary of the loaded corpus (entry count by category) and exits.

Run the tests

cd mcp-server
pip install -e ".[dev]"
pytest -q

The tests cover:

test_loader.py — the loader can find and parse every entry.
test_no_drm_names.py — the DRM-name denylist is clean (the project's content policy guard).
test_schema_sync.py — the Python Pydantic schema and the Astro Zod schema agree on field names and the category/platform/ difficulty enums.
test_search.py — the in-memory search index.
test_tools.py — every MCP tool returns what its description promises.

Publishing a new release

The repo ships a python-publish.yml workflow (.github/workflows/) that builds the sdist + wheel and publishes to PyPI on every GitHub release, using PyPI trusted publishing (OIDC). No API tokens are stored in GitHub secrets.

One-time PyPI setup (do this once before the first release):

Log in to https://pypi.org/ and create the re-library-mcp project if it doesn't already exist. PyPI project URLs are reserved on first upload; you don't need to "create" the project in advance.
Go to https://pypi.org/manage/account/publishing/ and add a new pending publisher with:
- PyPI Project Name: re-library-mcp
- Owner: Heretek-AI
- Repository name: RE-Library
- Workflow filename: python-publish.yml
- Environment name: pypi (must match the environment in the workflow file — create it under GitHub Settings → Environments)
(Optional) Repeat the same steps on https://test.pypi.org/ if you want a dry-run target.

To cut a release:

# 1. Bump version in pyproject.toml (and any version references)
# 2. Commit
# 3. Tag and push
git tag v0.1.0
git push --tags

# 4. Open the release on GitHub
gh release create v0.1.0 --generate-notes

# The release event triggers the publish workflow, which builds and
# uploads to PyPI. Watch progress under the Actions tab.

You can also trigger the workflow manually from the Actions tab (workflow_dispatch) — useful for a TestPyPI dry-run.

Development

The server is small and self-contained. Three modules do almost all the work:

loader.py — finds and parses entries (local dir or GitHub).
search.py — builds an in-memory inverted index.
tools.py — the five MCP tool implementations.

server.py is the thin glue that registers them with the mcp Python SDK over stdio.

The full content corpus is at ../content/ — add an entry there, restart the server, and it's available.

License

MIT. See ../LICENSE.