Skip to main content
Glama

mcpstat

PyPI - Version GitHub License PyPI - Python Version GitHub Actions Workflow Status PyPI - Downloads Checked with mypy Security: bandit Codecov

Python observability and usage analytics for MCP (Model Context Protocol) servers.

Track tool, prompt, and resource calls, successful and failed invocations, and execution latency in local SQLite storage. Add tracking to your existing Python MCP server and expose its statistics as MCP tools that an assistant can query. Optional file logging provides an audit trail.

Installation

pip install mcpstat
pip install "mcpstat[server]"  # Includes the SDK for the example below

Related MCP server: MCP Pulse

Quick Start

from mcp.server import MCPServer
from mcpstat import MCPStat, disable_sdk_tracing

app = MCPServer("my-server")
disable_sdk_tracing(app)
stat = MCPStat("my-server")


@app.tool()
@stat.track(name="echo")
async def echo(message: str) -> str:
    """Return a message, tracking usage and latency automatically."""
    return message


@app.tool()
async def get_tool_usage_stats() -> dict:
    """Return cumulative MCP usage statistics."""
    return await stat.get_stats()


if __name__ == "__main__":
    try:
        app.run()
    finally:
        stat.close()

Connect your assistant to this server, then ask: "Give me MCP usage stats"

For a complete request-to-statistics example, follow the weather server walkthrough. It demonstrates successful calls, an intentional failure, unused tools, and statistics that persist after restarting the server.

Choose Your Integration

Use case

Installation

What it tracks

Add tracking to an existing Python server

pip install mcpstat

The handlers you instrument with the SDK-independent library; keep your server's MCP SDK installed separately.

Run the bundled documentation and demonstration server

pip install "mcpstat[server]"

Its own temperature conversion tools and other registered handlers.

The decorator records usage; register a statistics tool such as get_tool_usage_stats to query that data from Claude Code, VS Code, or another MCP client. The standalone server does not automatically observe other MCP servers.

Features

  • SQLite-backed - Stats persist across restarts

  • Built-in MCP tools - get_tool_usage_stats, get_tool_catalog

  • Tag system - Categorize and filter tools

  • Token tracking - Estimate or record actual token usage

  • Latency tracking - Measure execution duration, identify slow tools

  • File logging - Optional timestamped audit trail

  • Async-first - Database operations run in worker threads with serialized transactions

Install mcpstat for the SDK-independent tracking library. The [server] extra includes the SDK and the runnable mcpstat documentation and demonstration server. The standalone server tracks its own calls; integrate the library into each server you want to monitor.

Documentation

Full Documentation - Quick start, API reference, examples

Examples

# Clone and run example server
git clone https://github.com/tekkidev/mcpstat.git
cd mcpstat
pip install -e ".[dev]"
python examples/example_server.py

See examples/ for minimal and full integration patterns.

Contributing

See CONTRIBUTING.md. Run tests with:

pip install "mcpstat[dev]"
pytest tests/

License

MIT - see LICENSE for details.


MCP is a trademark of its respective owners. This project is not affiliated with or endorsed by any trademark holders.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      D
      maintenance
      Enables sharing MCP servers with clients, storing tool calls in PostgreSQL, and providing usage analytics via an admin dashboard and API.
      -
    • A
      license
      A
      quality
      B
      maintenance
      An MCP server that records agent execution metrics and exposes a Context Window Explorer to visualize exactly what entered the model's context window across sessions, tokens, and tool calls.
      8
      3
      MIT
    • F
      license
      Not graded
      quality
      D
      maintenance
      Read-only MCP server for querying AI usage metrics (exact tokens, cost, latency, errors, productivity) from a local SQLite database, enabling charts and analysis in Claude Code and Cursor.
      -