effect-mcp-server

Build an MCP server with Effect by defining tools and parameters using Schema for validation, merging MCP part layers with Layer.mergeAll, and providing a transport layer to expose typed capabilities to LLM clients.

The Model Context Protocol (MCP) is a standard for exposing tools, resources, and prompts to LLM clients. You need an MCP server when integrating AI assistants with typed, runtime-safe capabilities for CLI deployments or web endpoints.

Yes, MCP servers support both stdio for desktop/CLI clients and HTTP for web/SSE-style deployments. Provide McpServer.layerStdio or McpServer.layerHttp, ensuring correct stderr logging configuration for the chosen transport.

Define parameterized resources in an Effect MCP server by using Effect Schema to model resource template parameters for validation and decoding. Merge these resource layers with toolkits and prompts using Layer.mergeAll before providing the transport layer.

The best way to structure an MCP server using the Effect layer pattern is to define tools, resources, and prompts as separate layers, merge them with Layer.mergeAll, and launch the combined layer as a long-lived process with transport-specific logging.

Stdio MCP server logging requires stderr because the stdio transport uses stdout for the actual MCP protocol communication with the LLM client. Logging to stdout would corrupt the protocol stream, so stderr is used for diagnostic output.