Colpali MCP Server - Python

What is Colpali MCP Server - Python?

Colpali MCP Server - Python is an MCP server for semantic image retrieval using ColPali and Elasticsearch. It indexes images or PDF pages and enables natural language search queries against the indexed content.

How to use Colpali MCP Server - Python?

Install by cloning the repository and running make install. Configure the server in your MCP client's mcp.json with the path to start_server.sh and environment variables for Elasticsearch (ES_HOST, ES_USER, ES_PASSWORD). The server exposes four tools: search_images, index_image, get_index_stats, and clear_index.

Key features of Colpali MCP Server - Python

• Semantic image search using natural language queries
• Automatic indexing of images and PDFs
• Multimodal embeddings combining vision and language (ColPali)
• Scalable storage with Elasticsearch
• Standard MCP API compatible with any MCP client

Use cases of Colpali MCP Server - Python

• Search for diagrams or schemas in a document database
• Index and retrieve screenshots by textual description
• Enable multimodal search across PDF archives
• Build AI‑powered image retrieval systems for RAG workflows

FAQ from Colpali MCP Server - Python

What dependencies are required?

Python 3.10+, Elasticsearch 8.0+, and a GPU is recommended for ColPali (though it also works on CPU).

Where are the image embeddings stored?

Embeddings and metadata are stored in Elasticsearch, accessed via the configured ES_HOST, ES_USER, and ES_PASSWORD.

How are images indexed?

Use the index_image tool to index images one by one (batch size of 1). PDFs can be indexed page by page by calling the tool for each page.

Is the clear_index tool reversible?

No, clear_index is irreversible and permanently removes all indexed data.

How is authentication handled?

Authentication is via the ES_USER and ES_PASSWORD environment variables passed to the server at startup.