ArcadeAI/arcade-mcp

What it solves

arcade-mcp provides a framework for developers to build custom Model Context Protocol (MCP) servers and tools. It solves the problem of integrating internal APIs, custom OAuth providers, and domain-specific tools into AI agents, ensuring they can securely interact with external services without exposing sensitive credentials to the LLM or client.

How it works

It uses a decorator-based Python API to define tools, resources, and prompts that comply with the MCP specification. For authentication, developers declare requirements (like GitHub scopes or API keys) within the code; the framework then handles the OAuth flow, token refreshing, and secure secret injection at runtime. The framework is vendor-neutral, meaning it works with any MCP-compliant client, LLM, or agent framework.

Who it’s for

This is for developers building AI agents or applications who need to connect their LLMs to private APIs, custom tools, or services requiring secure authentication that aren't available in prebuilt tool catalogs.

Highlights

• Decorator API covering the full MCP specification, including tools, resources, and logging.
• Secure authorized tool calling that keeps secrets and OAuth tokens hidden from the LLM and client.
• Built-in support for 23 popular service providers like GitHub, Slack, Notion, and Figma.
• Includes a CLI for scaffolding servers, configuring clients, and deploying to Arcade Cloud.
• Supports testing tool-call accuracy with built-in evaluation tools.