mcp-pulse 0.1.0

Drop-in observability for MCP servers — track tool calls, latency, errors, and usage with one line of code

mcp-pulse

Drop-in observability for MCP servers. Track tool calls, latency, errors, and usage with one line of code.

Install

pip install mcp-pulse

Quick Start

from mcp_pulse import ObserveMCP

# Replace FastMCP with ObserveMCP — that's it
mcp = ObserveMCP("my-server")

@mcp.tool()
def my_tool(param: str) -> str:
    """My tool description."""
    return f"result for {param}"

mcp.run(transport="stdio")

Every tool call is now automatically tracked — tool name, duration, success/failure, response size.

Dashboard

pip install mcp-pulse[dashboard]
mcp-pulse

Opens a web dashboard at http://localhost:8020 showing:

• Total calls, error rate, avg latency
• Calls per hour chart
• Per-tool breakdown (call count, p50/p95 latency, error rate)
• Recent call log

Alternative: Wrap an existing server

from mcp.server.fastmcp import FastMCP
from mcp_pulse import observe

mcp = FastMCP("my-server")

@mcp.tool()
def my_tool(param: str) -> str:
    return f"result for {param}"

observe(mcp)  # instruments all registered tools
mcp.run(transport="stdio")

Options

mcp = ObserveMCP(
    "my-server",
    db_path="/path/to/custom.db",  # default: ~/.mcp-pulse/observe.db
    log_params=True,               # log input parameters (default: False)
)

License

MIT