After successful connection, a Connection id will be returned, which can be used later to disconnect
Recommended LLM model are GPT-4o, Claude 3.5 Sonnet, other models may have limited support
Features
Supports MCP server features: Tools, Resources (without using the /autoContext command, a popup will appear requiring users to select which Resources to reference), and listening for respective list update events
Supports MCP client features: Roots and Sampling
Supports smart followups suggestions
Supports managing multiple MCP service connections
Supports actively disconnecting and reconnecting MCP services
Supports importing and connecting MCP services from local json files, which must follow this format:
/connectLocal - Connect to a local MCP service process
/connectRemote - Connect to a remote SSE MCP service. Note: remote service connections are currently unstable
/disconnect - Disconnect a specific MCP connection, use the connectionID that was returned after successful connection
@devlinker /disconnect connectionID
/disconnectAll - Disconnect all MCP connections
/refresh - Refresh all MCP service connections
/load - After dragging an MCP service configuration json file into the chat window as an attachment, use this command to load and connect to all services
/autoContext - Let the LLM choose which external resources to reference. Use with caution as there is a risk of the LLM using outdated data. Users need to explicitly request refreshing data to get the latest information
Common Issues
Before creating and connecting to a local MCP service process, it is recommended to first execute it through your local terminal to troubleshoot any issues
When the display shows 0 connected MCP services but disconnect and other followups still appear in the chat window, don't worry about it. The next conversation turn will refresh these options
If every conversation returns the same error and you can't resolve it, try starting a new chat