lokryn-mcp-log 0.2.0

Compliance-grade audit logging for MCP client operations

lokryn-mcp-log

Compliance-grade audit logging for MCP (Model Context Protocol) client operations.

What It Does

Wraps the official MCP Python SDK's ClientSession and logs every operation—tool calls, resource access, prompt executions—in a format that satisfies SOC2, HIPAA, and PCI audit requirements.

Logs conform to the lokryn-compliance-log-schema, an open standard for audit logging (schema vendored in this package, will be extracted to separate PyPI package).

Installation

pip install lokryn-mcp-log

Or with uv:

uv add lokryn-mcp-log

Quick Start

from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client
from lokryn_mcp_log import with_logging, StdoutSink

server_params = StdioServerParameters(command="python", args=["server.py"])

async def main():
    async with stdio_client(server_params) as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()

            # Wrap with logging
            logged = with_logging(
                session,
                sink=StdoutSink(),
                environment="development",
            )

            # Use exactly like normal
            await logged.call_tool("add", {"a": 1, "b": 2})

Sinks

StdoutSink

from lokryn_mcp_log import StdoutSink

sink = StdoutSink()           # JSON to stdout
sink = StdoutSink(pretty=True) # Pretty-printed

FileSink

from lokryn_mcp_log import FileSink

sink = FileSink("/var/log/mcp-audit.jsonl")

HTTPSink

from lokryn_mcp_log import HTTPSink

sink = HTTPSink(
    endpoint="https://your-log-collector.com/ingest",
    headers={"Authorization": "Bearer <token>"},
)

Custom Sink

Implement the Sink protocol:

from lokryn_mcp_log.schema import log_pb2

class MySink:
    async def emit(self, record: log_pb2.LogRequest) -> None:
        # Your logic here
        pass

Configuration

from lokryn_mcp_log import with_logging
from lokryn_mcp_log.schema import log_pb2

logged = with_logging(
    session,
    sink=my_sink,
    environment="production",           # Required
    actor_id="agent-001",                # Optional, auto-generated if omitted
    component="my-agent",                # Defaults to "mcp-client"
    policy_tags=["SOC2", "HIPAA"],       # Optional compliance tags
    default_sensitivity=log_pb2.SENSITIVITY_CONFIDENTIAL,
)

What Gets Logged

Operation	Event Type	Resource
initialize()	EVENT_LOGIN	session/initialize
list_tools()	EVENT_TOOL_INVOCATION	tools/list
call_tool(name, args)	EVENT_TOOL_INVOCATION	tools/{name}
list_resources()	EVENT_RESOURCE_ACCESS	resources/list
read_resource(uri)	EVENT_CONTEXT_ACCESS	{uri}
list_prompts()	EVENT_PROMPT_EXECUTION	prompts/list
get_prompt(name)	EVENT_PROMPT_EXECUTION	prompts/{name}
(session close)	EVENT_LOGOUT	session/close

Each log includes:

• Timestamp
• Actor ID (session/agent identifier)
• Duration (milliseconds)
• Correlation ID (for tracing)
• Input arguments
• Outcome (success/failure)
• Error details (on failure)

Error Handling

This is a library, not a service. If the sink fails, the exception propagates to your code. Handle it as appropriate for your use case.

Field Notes Integration

Send logs directly to Field Notes for tamper-evident storage and querying:

from lokryn_mcp_log import HTTPSink

sink = HTTPSink(
    endpoint="https://fieldnotes.lokryn.com/ingest",
    headers={"Authorization": "Bearer <your-api-key>"},
)

License

AGPL-3.0. Commercial license available—contact [e-mail].

Links

• lokryn-compliance-log-schema
• Field Notes
• MCP Python SDK