mcp-server AGENTS.md

AGENTS.md

This file provides guidance to AI coding agents when working with code in this repository.

Commands

# Install dependencies
uv sync

# Run the server
uv run python main.py

# Run all tests
uv run pytest

# Run a single test file
uv run pytest tests/tools/test_sum.py

# Run a single test by name
uv run pytest tests/test_integration.py::test_call_sum_tool

# Lint and format (runs Ruff)
uv run pre-commit run --all-files

# Install pre-commit hooks (one-time setup)
uv run pre-commit install

Architecture

This is a FastMCP server exposing scholarly tools via the Model Context Protocol.

• main.py is the default runtime entry point.
• server/app.py provides the app factory (create_app) and registers tools.
• server/core/ contains settings and transport normalization.
• server/services/ contains business logic used by tools.
• server/tools/ defines and mounts MCP tool sub-servers.
• server/utils/ contains runtime helpers (for transport-specific kwargs).

Tool sub-server pattern

Each tool lives in its own file under server/tools/ and creates its own FastMCP instance (a "sub-server"). Tools are registered via the @server.tool() decorator. The sub-server is then mounted onto the main app in server/tools/__init__.py's register_tools() function with a namespace.

• Tool names exposed to MCP clients follow the pattern {namespace}_{function_name} (e.g., sum_add from namespace "sum" and function add).
• Type hints define the JSON input schema; docstrings become the tool description visible to clients.

Adding a new tool

1. Create service logic in server/services/my_tool.py (business logic only).
2. Export the service from server/services/__init__.py.
3. Create server/tools/my_tool.py with a FastMCP("my_tool") sub-server and @server.tool() functions that call the service.
4. Mount it in server/tools/__init__.py: app.mount(my_tool_server, namespace="my_tool").
5. Add unit tests in tests/tools/test_my_tool.py; update tests/test_contract.py (tool count, metadata) and tests/test_integration.py.