Arcade MCP

Open-source Python framework for building MCP servers and tools.

Documentation • Examples • Discord

What is arcade-mcp

arcade-mcp is the Python framework for building Model Context Protocol servers and the tools that run inside them. It powers the 7,500+ prebuilt tools across 81 MCP servers at Arcade.dev, and is open-sourced so you can build your own.

Use it when you need MCP tools that aren't already in the prebuilt catalog: internal APIs, custom OAuth providers, domain-specific integrations.

Highlights:

• Decorator API covering the full MCP spec: tools, resources, prompts, sampling, elicitation, progress, logging.
• Authorized tool calling. Declare requires_auth=GitHub(scopes=["repo"]) and Arcade handles OAuth, token refresh, and per-call scoping. The client and the LLM never see the token.
• Vendor-neutral. Any MCP client, any LLM, any agent framework (LangChain, Mastra, Pydantic AI, CrewAI, Google ADK, OpenAI Agents).
• arcade evals for testing tool-call accuracy against real LLMs.
• arcade deploy for one-command hosting on Arcade Cloud.

Stateless tool elicitation

MCP 2026-07-28 HTTP connections support form and URL elicitation through tools/call. Ordinary resource and prompt operations remain available; elicitation through those operations is deferred.

Set ARCADE_WORKER_SECRET to the same value of at least 32 UTF-8 bytes on every replica that can resume a tool. Without it, continuations are process-local and cannot survive a restart. Short configured keys disable elicitation state signing; ordinary tool calls remain available. Startup logs report both configurations.

Retries replay the tool. Keep input-request ordering deterministic and make side effects before each input request idempotent.

Authorized tool calling

Tools can declare:

1. Which OAuth scopes they need, and Arcade handles the rest: prompting the end user to authorize, storing and refreshing tokens, scoping them per call
2. API keys or any other secrets, and Arcade hosts the secrets securely in an encrypted environment.

The secrets (OAuth tokens, API keys, etc) are securely injected by Arcade into your tool call at runtime, so that your tool can authorize requests to upstream APIs, for example. The client and the LLM never see the secret values.

A tool that reads the user's GitHub repos is one decorator away:

from arcade_mcp_server import MCPApp, Context
from arcade_mcp_server.auth import GitHub

app = MCPApp(name="gh", version="1.0.0")

@app.tool(requires_auth=GitHub(scopes=["repo"]))
async def list_my_repos(context: Context) -> list[str]:
    """List the authenticated user's GitHub repositories."""
    token = context.get_auth_token_or_empty()
    ...

When the tool is invoked through Arcade Cloud, the user is presented with a URL to complete the OAuth challenge in their browser. On success, the token is injected into context for that call. Subsequent calls reuse and refresh the token automatically.

23 helper classes ship with the framework for popular providers:

Asana, Atlassian, Attio, Calendly, ClickUp, Discord, Dropbox, Figma, GitHub, Google, Hubspot, Linear, LinkedIn, Microsoft, MicrosoftPowerBI, Notion, PagerDuty, Reddit, Slack, Spotify, Twitch, X, Zoom.

For any other OAuth API, use the generic OAuth2(...) class and register your OAuth app in the Arcade Dashboard.

Quick Start

Install the CLI

uv tool install arcade-mcp

Scaffold a new server

arcade new my_server
cd my_server/src/my_server

The scaffold creates pyproject.toml, .env.example, and a server.py with example tools.

A minimal tool:

from typing import Annotated
from arcade_mcp_server import MCPApp

app = MCPApp(name="my_server", version="1.0.0")

@app.tool
def greet(name: Annotated[str, "Name to greet"]) -> str:
    """Greet a person by name."""
    return f"Hello, {name}!"

if __name__ == "__main__":
    app.run(transport="stdio")

Run

uv run server.py             # stdio (default), for Claude Desktop and CLI tools
uv run server.py http        # HTTP+SSE, for Cursor and VS Code; docs at http://127.0.0.1:8000/docs

Configure your MCP client

arcade configure claude                                 # Claude Desktop, stdio
arcade configure cursor --transport http --port 8080    # Cursor, local HTTP on :8080
arcade configure vscode --entrypoint my_server.py       # VS Code, stdio launching my_server.py

For more patterns (MCP resources, sampling, progress reporting, tool chaining, end-to-end agents, eval suites, Resource Server Auth), browse examples/mcp_servers/. Run arcade --help for the full CLI.

With Arcade Cloud

arcade login followed by arcade deploy packages your server, discovers and upserts required secrets, and polls until it's healthy on Arcade Cloud. From there, the Arcade Engine fulfills authorized tool calling flows for end users, and arcade server logs/list/status plus arcade dashboard provide observability and management.

Standalone is supported too. Run your server in any MCP client over stdio or HTTP. Supply your own access tokens for tools with requires_auth=..., and protect production HTTP endpoints with Resource Server Auth (OAuth 2.1 Bearer tokens validated against your IdP).

Install from Source

Requires Python 3.10+ and uv.

git clone https://github.com/ArcadeAI/arcade-mcp.git
cd arcade-mcp
make install

See CONTRIBUTING.md for development workflow and SECURITY.md for vulnerability reporting.

Community

• Discord: https://discord.com/invite/GUZEMpEZ9p
• X: https://x.com/TryArcade
• LinkedIn: https://www.linkedin.com/company/arcade-mcp
• Issues: https://github.com/ArcadeAI/arcade-mcp/issues
• Documentation: https://docs.arcade.dev

License

MIT. See LICENSE.