ApeRAG

What apecloud/ApeRAG does

The apecloud/ApeRAG MCP server gives AI assistants access to an ApeRAG deployment and its document collections. ApeRAG is an end to end RAG platforms project that combines Graph RAG, vector retrieval, full-text search, summary-based retrieval, and vision-oriented search. Through MCP, an assistant can inspect available collections, search their contents, and submit natural-language questions about the documents.

The underlying platform is intended for teams building knowledge bases and agent applications. It includes a web interface, an API, document management, graph visualization, audit logging, LLM model management, and agent workflow management. These capabilities belong to the broader ApeRAG application; the MCP interface exposes the documented collection and query functions to connected assistants.

How it works

ApeRAG runs its MCP endpoint at /mcp/ on the application API. Clients connect using an HTTP URL and authenticate with a bearer token. The token can be sent in the Authorization header, which the documentation recommends, or supplied through APERAG_API_KEY as a fallback.

Search can use more than one retrieval method. Vector and full-text indexes support semantic and lexical matching, while Graph RAG uses relationships in the knowledge graph. Summary and vision indexes extend retrieval to condensed document content and visual material. Built-in AI agents can identify relevant collections and search content, but the MCP tools described for clients focus on collection browsing, hybrid search, and intelligent questions.

Setup and configuration

The documented local setup uses Docker Compose. A machine should have at least two CPU cores, 4 GiB of RAM, Docker, and Docker Compose. Clone the repository, copy envs/env.template to .env, and start the services with Docker Compose. The web interface is then available on port 3000, while API documentation is available on port 8000.

Configure an MCP client with the deployed API origin, the /mcp/ path, and a valid ApeRAG API key. Local deployments use http://localhost:8000/mcp/; a hosted deployment must use its own origin instead. Kubernetes deployment is also documented through a Helm chart. That route can use existing PostgreSQL, Redis, Qdrant, and Elasticsearch services or deploy them with KubeBlocks.

Tools and capabilities

The apecloud/ApeRAG MCP server supports:

• Listing and exploring knowledge collections.
• Searching with vector, full-text, and graph retrieval methods.
• Asking natural-language questions about indexed documents.
• Accessing a platform that supports multimodal document processing and vision analysis.

ApeRAG optionally integrates the MinerU-based document parsing service for complex documents, tables, formulas, and scientific content. Docker Compose profiles support standard or GPU-accelerated parsing. The platform also includes entity normalization for graph construction and merging related entities.

Limitations and notes

The MCP endpoint requires a running ApeRAG deployment and a valid API key. Enhanced parsing is optional and adds a separate service; the Kubernetes documentation notes that the default doc-ray service needs at least four CPU cores and 8 GB of RAM. Database services are also required for the Kubernetes deployment, including PostgreSQL, Redis, Qdrant, and Elasticsearch.

The apecloud/ApeRAG MCP server does not describe client-specific setup for any particular assistant. Client compatibility therefore depends on support for authenticated HTTP MCP connections. The repository excerpt also does not specify a license name.