GEMINI-MCP
Repository intelligence MCP server
Advanced Model Context Protocol server exposing repository AST exploration, LanceDB semantic vector search, and Gemini architectural mapping to coding agents.
subjectEXECUTIVE SUMMARY
Large-scale software repositories present significant context-window challenges for coding agents. Agents often struggle to locate relevant files, misinterpret architectural dependencies, or generate ungrounded code patches that cause regressions.
Gemini-MCP bridges Google Gemini models with the local filesystem through standardized MCP tools: agents can generate structured multi-step implementation plans (plan_task), review diffs for security and logic flaws (review_patch), and semantically discover related code using local LanceDB vector embeddings (semantic_search_code).
Additionally, the server generates functional signatures and import dependency graphs (map_architecture), which can be inspected directly in a built-in web visualization UI at http://localhost:8000. All tool invocations and reasoning steps are recorded in a persistent SQLite audit log (history.db).
Coding agents lack semantic understanding of large codebases, frequently guessing file locations or missing structural dependencies. Standard text search fails when terminology differs between user queries and source code.
Gemini-MCP equips agents with a standardized tool suite featuring local LanceDB vector search, AST dependency graphing, and Gemini-powered architectural reasoning.