If you're working with Firestore and tired of Claude inventing field names or guessing at schema structure, this MCP server pulls your actual collections, field types, and presence rates directly into context. It exposes tools that let the AI query your live schema before writing database code, so it stops hallucinating `userId` when the field is actually `user_id`. Built on top of the LintBase CLI, which samples documents to infer schema and catch drift. Useful when you're prototyping queries, refactoring data models, or just want your agent to know what fields actually exist in production without manually pasting schema docs into every conversation.
claude mcp add --transport stdio io.github.madia333-lintbase-mcp uvx lintbase-mcp