ArcadeAI/arcade-mcp

What it solves

Arcade MCP provides a Python framework to build servers and tools that follow the Model Context Protocol (MCP). It simplifies the creation of custom integrations for internal APIs, domain-specific tools, and services requiring complex authentication, allowing these tools to be used by any MCP-compatible client or LLM.

How it works

The framework uses a decorator-based API to define tools, resources, prompts, and other MCP specifications. It supports both stdio and HTTP+SSE transports for compatibility with clients like Claude Desktop, Cursor, and VS Code. A key feature is its "Authorized Tool Calling" system, which handles OAuth flows, token refresh, and secret management securely. When a tool requires authentication, Arcade manages the user authorization process and injects the token into the tool's context at runtime, ensuring the LLM and the client never see the secret values.

Who it’s for

Developers building AI agents and LLM applications who need to create custom, secure tool-sets for their models to interact with external APIs and data sources.

Highlights

• Full MCP Spec Support: Covers tools, resources, prompts, sampling, elicitation, and progress reporting.
• Secure Authentication: Built-in helper classes for 22 popular OAuth providers (e.g., GitHub, Google, Slack) and support for generic OAuth2.
• Vendor Neutral: Works with any MCP client, LLM, or agent framework like LangChain, CrewAI, or Pydantic AI.
• Developer Tooling: Includes arcade evals for testing tool-call accuracy and arcade deploy for one-command hosting on Arcade Cloud.
• Flexible Deployment: Supports both local execution via stdio/HTTP and cloud hosting.