CIViC MCP Server

Overview

The CIViC MCP Server transforms the wealth of cancer‑variant knowledge stored in the Clinical Interpretation of Variants in Cancer (CIViC) database into a format that AI assistants can interrogate directly. By running on Cloudflare Workers, the server offers low‑latency access while leveraging Durable Objects to cache and process GraphQL responses as SQLite tables. This architecture bridges the gap between a complex, crowd‑sourced API and the structured data expectations of modern AI tools.

At its core, the server exposes two primary tool families. The GraphQL Query Tool fetches raw CIViC data, while the SQL Query Tool operates on a local SQLite snapshot of that data. The GraphQL tool is marked as openWorldHint, indicating it reaches out to the external CIViC API, whereas the SQL tool is readOnlyHint and idempotent, ensuring safe, repeatable queries. This dual‑layer approach lets developers write natural‑language prompts that first gather data and then perform analytical SQL queries without leaving the assistant’s context.

Key capabilities include structured tool output with rich fields, allowing the assistant to present results in user‑friendly tables or charts. The server also supports protocol version headers and human‑readable titles, which improve the usability of tool responses in conversational UI. Error handling follows MCP best practices, returning clear, structured messages that can be displayed or logged by the client.

Developers can integrate this server into AI workflows in several ways. For example, a clinical decision‑support system could ask an assistant to “list all CIViC variants associated with BRCA1 that have a high evidence level,” triggering the GraphQL tool to fetch relevant entries and the SQL tool to filter by evidence score. In research pipelines, a bioinformatician could query variant‑driven drug response data across multiple cohorts by composing SQL queries that run against the cached SQLite tables, all without writing custom API clients.

The CIViC MCP Server offers unique advantages: it eliminates the need for developers to build their own GraphQL‑to‑SQL translators, provides a standardized, secure interface aligned with the latest MCP specifications, and scales automatically through Cloudflare Workers. While some features such as streamable HTTP transport and OAuth 2.1 authorization are still pending, the current implementation already delivers a powerful, developer‑friendly bridge between AI assistants and cancer genomics data.