Skip to main content
Glama
Agent-Hellboy

PyMCP Kit

PyMCP Kit

Python CI codecov MCP Conformance-v2025-11-25 MCP Protocol Revisions PyPI - Version PyPI Downloads Docs License: MIT Python 3.11+

Documentation | Quick Start | Tasks | Security | Middleware

PyMCP Kit is a capability-first MCP server toolkit for FastAPI. It keeps the built-in transport surface small, supports Streamable HTTP and stdio, and ships app-scoped registries, roots, tasks, optional auth hooks, OAuth protected-resource metadata, and opt-in draft MCP support without pulling in a larger framework.

Quick Start

Install from PyPI:

pip install pymcp-kit

For local development from this repo:

pip install -e .

Register tools, prompts, and resources, then build an app:

from pymcp import (
    CapabilitySettings,
    ServerSettings,
    create_app,
    prompt_registry,
    resource_registry,
    tool_registry,
)


@tool_registry.register
def add(a: float, b: float) -> str:
    return str(a + b)


@prompt_registry.register(description="Create a release summary prompt.")
def summarize_release(topic: str) -> str:
    return f"Summarize the release impact for {topic}."


@resource_registry.register(
    uri="memo://release-plan",
    name="release_plan",
    description="Latest release checklist",
    mime_type="text/markdown",
)
def release_plan() -> str:
    return "# Release Plan\n- freeze API\n- tag build\n"


@resource_registry.register_template(
    uri_template="note://{topic}",
    name="topic_note",
    description="Parameterized note resource keyed by topic.",
)
def topic_note(topic: str) -> str:
    return f"Notes for topic: {topic}"


app = create_app(
    server_settings=ServerSettings(
        name="demo-server",
        version="0.1.0",
        capabilities=CapabilitySettings(
            advertise_empty_prompts=False,
            advertise_empty_resources=False,
        ),
    )
)

The HTTP transport is mounted at /mcp. For local-process integrations, use run_stdio_server(app).

Stable MCP revisions are enabled by default. To build against the draft stateless revision, opt in explicitly:

app = create_app(protocol_mode="draft")  # 2026-07-28 only
app = create_app(protocol_mode="dual")   # draft plus stable clients

Draft extension capabilities are advertised under the spec extensions object. Built-in task support becomes extensions["io.modelcontextprotocol/tasks"] for the draft revision, and additional namespaced extensions can be supplied with CapabilitySettings(extensions={...}).

Official extension capability advertisement is opt-in:

from pymcp import CapabilitySettings, ServerSettings, create_app

app = create_app(
    server_settings=ServerSettings(
        protocol_mode="draft",
        capabilities=CapabilitySettings(
            mcp_apps_enabled=True,
            oauth_client_credentials_enabled=True,
            enterprise_managed_authorization_enabled=True,
        ),
    )
)

Hosted documentation is built from docs/ with MkDocs Material and published to GitHub Pages.

Related MCP server: Production MCP Template

Features

  • Streamable HTTP transport for networked MCP servers

  • Stdio transport for local-process MCP hosts

  • Opt-in draft 2026-07-28 stateless protocol support with server/discover and namespaced extensions

  • Tool, prompt, and resource registries, including parameterized resource templates

  • Roots, resource subscriptions, and app-scoped session lifecycle

  • Task-aware tool execution with progress, cancellation, and result polling

  • Optional authentication and authorization hooks with OAuth protected-resource metadata discovery

  • Capability advertising through CapabilitySettings

  • FastAPI middleware integration through MiddlewareConfig

  • Small surface area focused on practical MCP server builds

Supported MCP Methods

Server-side JSON-RPC methods and notifications implemented by pymcp-kit:

Lifecycle

  • initialize

  • ping

  • notifications/initialized

  • notifications/cancelled

  • server/discover (draft 2026-07-28, when protocol_mode="draft" or "dual" is enabled)

Tools

  • tools/list (cursor pagination)

  • tools/call

Prompts

  • prompts/list (cursor pagination)

  • prompts/get

Resources

  • resources/list (cursor pagination)

  • resources/templates/list (cursor pagination)

  • resources/read

  • resources/subscribe

  • resources/unsubscribe

  • notifications/resources/updated (server → client)

  • notifications/resources/list_changed (server → client, when enabled)

Completions

  • completion/complete (when completions_enabled is set)

Tasks

  • tasks/list (cursor pagination)

  • tasks/get

  • tasks/cancel

  • tasks/result

  • notifications/tasks/status (server → client)

  • notifications/progress (server → client)

Client capabilities (server-initiated helpers)

These are not inbound server handlers; the toolkit sends requests or notifications to the client when the client advertises the capability:

  • roots/list via request_roots_list()

  • notifications/roots/list_changed (client → server notification)

  • elicitation/create via request_elicitation()

  • sampling/createMessage via request_sampling()

  • notifications/message via send_log_message()

List operations support optional cursor / nextCursor pagination. Page size is controlled by CapabilitySettings.list_page_size (default 50).

See Runtime Surface for protocol versions, HTTP endpoints, and capability settings.

Example Server

Run the bundled example server:

python example/run_server.py

That starts a FastAPI app on http://127.0.0.1:8088 with the MCP endpoint mounted at http://127.0.0.1:8088/mcp.

Stdio Transport

from pymcp import create_app, run_stdio_server


app = create_app()
run_stdio_server(app)

Middleware

Middleware stays separate from capability registration. Use MiddlewareConfig to control CORS, compression, logging, auth hooks, and custom ASGI middleware, then pass it into create_app(). See the hosted Middleware page for examples.

Scope

  • Prompts and resources are advertised only when registered by default

  • Registries are copied into an app-scoped manager when create_app() runs

  • Streamable HTTP and stdio are the only built-in transports

  • Extra transports such as SSE and HTTP NDJSON are intentionally not shipped in pymcp-kit

A
license - permissive license
-
quality - not tested
A
maintenance

Maintenance

Maintainers
6dResponse time
8wRelease cycle
3Releases (12mo)
Commit activity

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Servers

  • A
    license
    A
    quality
    D
    maintenance
    A production-grade, extensible Python template for building Model Context Protocol servers with support for Streamable HTTP and stdio transports. It provides a structured framework for implementing tools, resources, and prompts with built-in authentication, observability, and background task management.
    Last updated
    11
    MIT
  • A
    license
    -
    quality
    D
    maintenance
    A FastAPI MCP server that uses Server-Sent Events as a transport layer, providing echo functionality through tools, resources, and prompts.
    Last updated
    12
    MIT

View all related MCP servers

Related MCP Connectors

  • AI Reasoning Cache & Consensus Layer with 11 MCP tools via Streamable HTTP.

  • MCP server for AI agents to plan, verify, and deploy Cloudflare-native apps.

  • Local-first RAG engine with MCP server for AI agent integration.

View all MCP Connectors

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/Agent-Hellboy/py-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server