app

Memgraph Codebase Helper

Browser chat app that wraps the Claude Agent SDK (same runtime as Claude Code) and points it at local clones of the Memgraph org's main repos:

• memgraph/memgraph — core C++ database
• memgraph/helm-charts — Kubernetes deployment
• memgraph/ai-toolkit — GraphRAG / LangChain / MCP / LlamaIndex integrations

Ask it anything about how any of them works — the agent uses Read, Grep, and Glob against the checkouts to answer with concrete file / line references, just like Claude Code does in a terminal.

Layout

app/
├── .env            # your local ANTHROPIC_API_KEY (copy from .env.example)
├── backend/
│   ├── main.py             # FastAPI + SSE streaming + agent runtime
│   ├── requirements.txt
│   └── Dockerfile
└── frontend/
    ├── index.html          # chat UI (vanilla + Tailwind CDN)
    ├── app.js
    ├── styles.css
    └── nginx.conf

On every startup the backend clones any missing repos and pulls each one to the tip of its tracked branch, so the agent always sees fresh code. Checkouts live in a persistent Docker volume (memgraph-repos) under Docker, or in backend/repos/ when run locally.

Setup

cd agents/memgraph-codebase-helper/app
cp .env.example .env      # then add your ANTHROPIC_API_KEY

Run with Docker (recommended)

cd agents/memgraph-codebase-helper/app
docker compose up --build

Open http://localhost:3001. The frontend service (nginx) serves the UI and proxies /api/* to the backend service (FastAPI + Claude Agent SDK). First startup clones all three repos into a named volume (memgraph-repos); every subsequent startup just pulls the latest on each tracked branch.

To stop: docker compose down. To wipe the checkouts too: docker compose down -v.

Run locally (without Docker)

cd agents/memgraph-codebase-helper/app
python3 -m venv .venv
source .venv/bin/activate
pip install -r backend/requirements.txt

# claude-agent-sdk spawns the Claude Code CLI under the hood, install it once:
npm install -g @anthropic-ai/claude-code

python3 backend/main.py

Open http://localhost:8001 — in dev the FastAPI server serves both the API and the static frontend on the same port. Repos are cloned into backend/repos/ on first run and pulled on every subsequent startup.

How it works

• On startup, backend/main.py makes sure every repo listed in REPOS is present in REPO_ROOT, cloning if missing and otherwise hard-resetting to origin/<branch> so the code is always current.
• The agent is created with REPO_ROOT as its cwd and with Read, Grep, Glob, and Bash tools enabled — the standard set Claude Code uses for codebase exploration. Because cwd is the parent of all three repos, a single Grep can search across them.
• Each chat session is a ClaudeSDKClient kept in memory (keyed by a browser-side UUID) so follow-up questions retain context. "New chat" resets it.
• Responses stream to the browser over Server-Sent Events. Text, tool calls, and tool results are rendered as separate UI elements.

Refreshing the checkouts

Restart the backend — docker compose restart backend or re-running python3 backend/main.py — and every repo is pulled to HEAD before the server comes up.