Repo2Graph MCP gives coding agents cited, token-bounded answers about a codebase. It indexes calls, imports and git co-change links, then returns the source blocks that answer a question.
PluginContact for pricing
Step 1: Install uv so that uvx is available.
Step 2: Optionally run uvx repo2graph demo to see five questions answered on a bundled example repo.
Step 3: In Claude Code, run claude mcp add repo2graph -- uvx --from "repo2graph[mcp]" repo2graph-mcp . from your repository.
Step 4: For Cursor, Claude Desktop and other clients, follow docs/mcp.md in the repository.
Step 5: Ask the agent where something is enforced, what calls a function, or what a change would affect.
No. Building, querying and the MCP server need no model, API key or database. Only opt-in paths such as rag --answer send the context pack to an LLM.
The README says it wins on cross-file structural questions and loses on single-file lexical ones, where grep is the better tool.
Packs are clamped under a token ceiling, 12k tokens max over MCP.
The README shows Claude Code setup and points to docs/mcp.md for Cursor, Claude Desktop and other clients.
Requirements
CodeCartographer MCPCodeCartographer turns a repository into layered architecture notes, behavioral contracts, defect findings and a language-agnostic reimplementation spec. It ships as a Pi extension, an MCP server and a drop-in template.
CodeIndex MCPcodeindex is an MCP server that indexes a codebase with tree-sitter and answers structural questions such as symbol definitions, callers and change impact. It exposes 16 tools to AI coding agents.
Neruva Agent MCPNeruva is a memory layer for coding agents, delivered as an MCP server backed by a hosted API. Agents recall earlier decisions, log what they do and leave handoff notes between sessions.