CodeAtlas Platform

AI-powered codebase intelligence platform — MCP Server, AST analysis, Knowledge Graph, and semantic memory with Oracle 26ai.

Ship a secure, multi-tenant codebase intelligence backend without rebuilding authentication, MCP tooling, semantic memory, and knowledge graph infrastructure from scratch. CodeAtlas Platform is an open-source foundation for developers who want a clear starting point for AI-native code analysis services.

Why this platform?

Related MCP server: RAG MCP Server

Architecture

AI IDE (Claude/Cursor) → MCP (stdio/SSE) → Platform :3381 → Oracle 26ai + Firebase + NVIDIA

Architecture diagrams

Quick start

Requirements

• Node.js 20+

• pnpm 9+ (corepack enable && corepack prepare pnpm@9 --activate)

• Oracle 26ai (Autonomous Database or self-hosted)

• Firebase service account (for multi-tenant auth)

• NVIDIA NIM API key (for embeddings)

1. Clone and install

git clone https://github.com/giauphan/codeatlas-platform.git
cd codeatlas-platform
pnpm install

2. Configure environment

cp .env.example .env
# Edit .env with your Oracle, Firebase, and NVIDIA credentials

See docs/CONFIGURATION.md or docs/DEVELOPMENT.md for the full env var reference.

3. Build

pnpm run build

4. Initialize database

pnpm run db-init

5. Start the server

# Production mode (SSE on :3381)
PORT=3381 pnpm start

# Development mode (hot reload)
pnpm run dev

# Stdio mode (for Claude Desktop — leave PORT unset)
pnpm start

Server runs at http://localhost:3381. Health check: GET /health.

MCP integration

Claude Desktop (stdio)

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "codeatlas": {
      "command": "npx",
      "args": ["-y", "codeatlas-ai"],
      "env": {
        "CODEATLAS_API_KEY": "your_api_key"
      }
    }
  }
}

Cursor / VSCode (SSE)

{
  "mcpServers": {
    "codeatlas": {
      "url": "http://localhost:3381/sse"
    }
  }
}

See docs/API_EXAMPLES.md for full curl flows and transport modes.

MCP tools (30+)

Full tool reference: docs/architecture/mcp.md.

Documentation

Known limitations

• Oracle dependency: Requires Oracle 26ai with VECTOR support. Not portable to PostgreSQL/MySQL without migration.

• Firebase auth: Multi-tenant mode requires Firebase Admin SDK + service account. API-key-only mode supported for single-tenant.

• NVIDIA embeddings: Vector search depends on NVIDIA NIM API. Without NVIDIA_API_KEY, queries fall back to date-ordered results.

• Local indexing: Pure cloud deployments cannot index local code — run the codeatlas-ai client locally to sync AST data.

• Oracle Instant Client: Thick mode requires downloading Oracle Instant Client separately (not bundled due to license).

• Dashboard: Management UI ships separately in dashboard/ — build and deploy independently.

See CHANGELOG.md for release history.

Contributing

Bug reports, documentation fixes, tests, and focused feature contributions are welcome. Start with CONTRIBUTING.md, then look for issues labeled good first issue or help wanted.

For security vulnerabilities, follow SECURITY.md instead of opening a public issue.

License

Distributed under the MIT License. Maintained by @giauphan.