Malloy MCP Server

The Malloy MCP Server bridges the gap between AI assistants and Malloy’s powerful data modeling capabilities. By exposing a full MCP interface, it lets tools such as Claude execute Malloy queries and retrieve metadata about projects, packages, and models without leaving the assistant’s environment. This eliminates the need for developers to manually spin up separate query engines or write custom adapters, enabling rapid integration of analytical logic into conversational workflows.

At its core, the server implements three primary functions: query execution, metadata retrieval, and robust error reporting. A dedicated MCP tool receives a Malloy query string, the target model path, and optional parameters, then forwards the request to the Malloy Publisher API. The result is streamed back as a structured JSON payload that can be directly consumed by the assistant. Simultaneously, resource endpoints such as expose project and package information, allowing the assistant to introspect data schemas or discover available models on demand. This unified API surface lets developers write concise tool calls that cover both data extraction and discovery.

Key capabilities include:

• Type‑safe implementation using Pydantic models, ensuring that query inputs and responses adhere to expected schemas.
• Comprehensive error handling: custom exceptions bundle message, context, and stack information so that the assistant can surface meaningful diagnostics to users.
• Extensive test coverage guarantees reliability across edge cases, from malformed queries to network failures with the publisher API.
• FastMCP‑based server: built on a lightweight, high‑performance framework that keeps latency low for real‑time interactions.

Typical use cases span from building data‑driven chatbots that can answer business questions on the fly, to automating report generation where an assistant composes a query, executes it through MCP, and formats the results. In data‑engineering pipelines, developers can embed the server into CI/CD workflows to validate model changes against live datasets. The ability to fetch metadata on demand also supports dynamic UI generation, where an assistant can present users with a list of available reports or datasets without hard‑coding options.

Because the server is fully compliant with MCP specifications, it plugs seamlessly into existing AI assistant stacks. Developers simply register the Malloy query tool and resource endpoints, then invoke them via standard tool‑call syntax. The server’s design prioritizes clarity and reliability, making it an attractive choice for teams that need to expose complex analytical logic to conversational agents without compromising on type safety or error transparency.