RepoBrain MCP server
Commit Nov 12, 2026Lets an AI client ask grounded questions about a repository and get answers with file paths and line numbers.
Also known as study8677/repobrain
- Stars
- 1,362
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 12, 2026
- License
- MIT
What is RepoBrain MCP?
RepoBrain lets an AI client ask grounded questions about a repository and get answers with file paths and line numbers.
Summary generated from the project's README
About RepoBrain MCP server
Developers seeking to understand unfamiliar codebases can use RepoBrain to perform grounded questions about their repositories. It functions as a knowledge engine that provides answers containing specific file paths and line numbers, ensuring the AI assistant remains rooted in the actual source code. The system works by using a multi-agent cluster to autonomously read modules and generate a local knowledge base stored in a .repobrain/ directory.
Users can set up the tool using a Python 3.10+ environment and connect it to an OpenAI-compatible API or a headless CLI like Trae. The workflow involves running rb-refresh to build the knowledge layer and rb-ask to query it. Because the knowledge is stored locally within the project, it can be shared across different IDEs and agents like Cursor, Windsurf, or VS Code with Copilot, preventing vendor lock-in and ensuring consistent context regardless of the host environment.
What you can do with it
- 1Trace implementation detailsAsk the AI to find where a specific API is implemented and identify all its callers.
- 2Analyze data flowRequest an explanation of how data moves from the user interface through to the backend systems.
- 3Assess change impactDetermine which specific modules will be affected if a certain function is modified.
- 4Onboard to new projectsUse automated agents to learn the architecture of a new codebase quickly.
Written from the project's README. Every use case is checked against a quote from it.
Automated checks
Ran Nov 13, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work
- Source repository linked and reachablestudy8677/repobrain is public and reachablegithub.com/study8677/repobrain ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-12github.com/study8677/repobrain ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubRepoBrain
Cross-IDE repository knowledge engine for grounded codebase Q&A.
Formerly known as Antigravity Workspace Template — same project, new name.
RepoBrain helps you understand an unfamiliar repository, trace an implementation,
and ask questions with source-file references. It keeps project knowledge in
.repobrain/, shared across supported IDEs and agents.
Project Philosophy
Cross-IDE repository knowledge engine for grounded codebase Q&A. Same .repobrain/ knowledge layer reads in every IDE; one engine, every host.
An AI Agent's capability ceiling = the quality of context it can read.
rb-refresh deploys a multi-agent cluster that autonomously reads your code — each module gets its own Agent that generates a knowledge doc. rb-ask routes questions to the right Agent, grounded in real code with file paths and line numbers.
Instead of handing Claude Code / Codex a repo-wide grep and making it hunt on its own, give it a ChatGPT for your repository.
Traditional approach: RepoBrain approach:
CLAUDE.md = 5000 lines of docs Claude Code calls ask_project("how does auth work?")
Agent reads it all, forgets most Router → ModuleAgent reads actual source, returns exact answer
Hallucination rate stays high Grounded in real code, file paths, and git history
| Problem | Without RepoBrain | With RepoBrain |
|---|---|---|
| Agent forgets coding style | Repeats the same corrections | Reads .repobrain/conventions.md — gets it right the first time |
| Onboarding a new codebase | Agent guesses at architecture | rb-refresh → ModuleAgents self-learn each module |
| Switching between IDEs | Different rules everywhere | One .repobrain/ folder — every IDE reads it |
| Asking "how does X work?" | Agent reads random files | ask_project MCP → Router routes to the responsible ModuleAgent |
Architecture is files + a live Q&A engine, not plugins. Portable across any IDE, any LLM, zero vendor lock-in.
Read the full project philosophy.
Quick Start
Ask an AI assistant that can run commands in your project to install RepoBrain:
Read AI_INSTALL.md and follow it to install RepoBrain in this project.
The guide installs the tools, configures a model backend, then builds the knowledge base and checks an answer. A logged-in headless CLI such as Trae can serve as the backend without an API key; an OpenAI-compatible API is also supported. Requirements: Python 3.10+, Git with at least one commit, and a clean worktree before refresh.
For platform-specific installation or manual configuration, see
the installation guide and the quick-start guide.
Installing RepoBrain into an existing project does not require rb init or rb-init.
Everyday Use
Run these commands in your project directory after installation and configuration:
rb-refresh
rb-ask "How does authentication work in this project?"
rb-refresh automatically builds the first knowledge base, updates affected groups
after committed changes, or continues a matching interrupted task. If knowledge is
already current, it skips without calling a model. After a failure, run the same
command again to continue.
Use this only when you want to regenerate all knowledge:
rb-refresh --full
rb-ask reads the existing knowledge base and answers with source evidence. It
reminds you when committed code is newer; it does not refresh during a question.
Examples of useful questions:
- Where is this API implemented, and what calls it?
- How does data move from the UI to the backend?
- Which modules are affected by changing this function?
How It Works
Source code → rb-refresh → .repobrain/ knowledge → rb-ask → answer with code references
RepoBrain groups related code, generates module knowledge, and routes each question to relevant context. Knowledge lives in your project, so supported IDEs share the same base. Updates are staged separately and become active only when complete; a failed update preserves the existing usable generation.
The first build can take several minutes or longer on large repositories and uses your configured model. Commit or stash local changes before refreshing. IDE rules and new-project scaffolding are optional; see the usage reference.
Support Matrix
| Use | Environments | How to connect |
|---|---|---|
| Native plugins | Claude Code, Codex CLI | Setup, refresh, and ask via slash commands. |
| Compatible IDEs | Cursor, Windsurf, Gemini CLI, VS Code + Copilot, Cline, Aider, DeepSeek Harness | Shared context files, CLI, or an MCP client. |
| Other agents and scripts | Any host that can run commands or call MCP tools | CLI/JSON output or optional rb-mcp. |
Platform-specific commands, JSON output, and MCP registration are documented in the usage reference and INSTALL.md.
Evaluation
A locked comparison on Flask, ripgrep, Vite, and Prometheus used the same source access and model route for RepoBrain and CodeGraph + Trae. The full report includes accuracy, query time, tokens, and cold-build cost. Results apply to that experiment; AI knowledge generation and static graph indexing are different workloads.
Documentation
- Installation and troubleshooting
- Commands, JSON, and optional integrations
- Knowledge generation and Q&A architecture
- External MCP tools · Sandbox configuration
- Release changes
- Full documentation: English · 中文 · Español
NLPM Audit Feedback — Improved by NLPM, a natural-language programming linter by xiaolai.
Contributing
Ideas are contributions too! Open an issue to report bugs, suggest features, or propose architecture.
Contributors
Star History
License
MIT License. See LICENSE for details.
Built for the AI-native development era
Friendly Link: LINUX DO
RepoBrain MCP questions
What does RepoBrain do?
It builds a knowledge base of your repository and routes questions to module agents so answers are grounded in real source code with file paths and line numbers.
Which IDEs and AI clients does RepoBrain support?
It works with Claude Code, Codex CLI, Cursor, Windsurf, Gemini CLI, VS Code with Copilot, Cline, Aider, and other hosts that can run commands or call MCP tools.
Does RepoBrain need an API key?
Not necessarily — a logged-in headless CLI can serve as the model backend, or you can use an OpenAI-compatible API.
How do I refresh the knowledge base?
Run rb-refresh, which builds the first knowledge base, updates affected groups after committed changes, or resumes an interrupted task; use --full to regenerate everything.
Where is the knowledge stored?
Project knowledge is kept in a .repobrain/ folder inside your repository and shared across supported IDEs and agents.