Skip to content

MCP Integration Overview

ChatShell supports the Model Context Protocol (MCP), an open standard for connecting AI applications to external tool servers. MCP allows you to extend ChatShell’s capabilities far beyond the 9 built-in tools.

MCP is a protocol that defines how AI applications discover and call tools provided by external servers. A single MCP server can expose dozens of tools — database access, file system operations, API integrations, custom business logic, and more.

ChatShell uses the rmcp Rust SDK to connect to MCP servers.

ChatShell supports two MCP transport modes:

TransportDescriptionUse case
HTTP (Streamable HTTP)Connect to a remote MCP server via HTTPHosted services, cloud tools
STDIORun a local MCP server as a child processLocal tools, npx-based servers

MCP servers are managed in Settings → MCP Servers.

From there you can:

  • Add new servers (HTTP or STDIO)
  • Paste JSON config from Claude Desktop, Cursor, or other tools to import servers instantly
  • Test connections before enabling
  • View available tools from each server
  • Delete servers you no longer need

When adding a new server, click Paste JSON config to switch to a JSON input view. ChatShell accepts several formats:

  • Standard mcpServers wrapper — the full { "mcpServers": { "name": { ... } } } format used by Claude Desktop and Cursor
  • Bare STDIO object{ "command": "npx", "args": [...] }
  • Bare HTTP object{ "url": "https://..." }

Paste the JSON, click Apply, and all form fields are filled in automatically.

MCP servers can be enabled or disabled per conversation. Click the Plug icon in the chat input toolbar to select which MCP servers are active for the current conversation.

This lets you give the AI only the tools it needs for a specific task, without cluttering every conversation with all available tools.

For MCP servers that require authentication, ChatShell supports OAuth 2.0 and 2.1 with PKCE. This includes enterprise MCP services that require user authentication before granting tool access.

The MCP Everything Server is a good starting point for testing. Add it via STDIO:

Command: npx
Args: -y @modelcontextprotocol/server-everything

This gives the AI access to a range of sample tools for testing MCP connectivity.