Compoid MCP

What Compoid MCP does

Compoid MCP MCP server exposes Compoid repository operations through the Model Context Protocol. It gives an AI client access to records, communities, and collection structures containing research, images, videos, papers, datasets, and other resource types.

Record search accepts a required query and optional filters for title, description, community, keywords, creators, publication dates, access status, resource type, file type, and sort order. Results include identifiers, descriptions, authors, dates, topics, URLs, and image previews. Community search provides comparable discovery for community names and metadata.

Detailed lookup tools retrieve a complete record by Compoid ID or OAI identifier, or a community by its ID. Open-access records can be downloaded as zip archives. Files can also be uploaded through a data URI when creating or updating repository content.

How it works

The hosted endpoint is https://mcpv.compoid.com/mcp and uses Streamable HTTP. Clients authenticate with the X-Compoid-Repo-Key header. The repository also publishes a Python package that runs as a local stdio MCP process, allowing clients to start compoid-mcp directly.

Compoid MCP MCP server passes tool requests to Compoid services and returns structured repository information to the connected model. Create and update operations can supply metadata and files. Record creation requires non-empty creator and keyword lists; subject values can include up to five display names from Compoid’s subject vocabulary, with “Artificial Intelligence” appended by default.

Setup and configuration

For a hosted connection, configure the MCP URL and replace YOUR_API_KEY with a Compoid repository key. The README documents setup for Claude Code, Cursor, ChatGPT Developer Mode, Claude Desktop, and VSCode Copilot.

For local use, install the package with pip install compoid-mcp or run it with uvx compoid-mcp. The local example configures repository, AI, upload, logging, sorting, download, and archive-extraction settings. Python 3.12 or later is required. The default AI model in the current configuration is Qwen.

Tools and capabilities

The documented tool set includes:

• Compoid_search_records for filtered record discovery.
• Compoid_search_communities for finding communities.
• Compoid_get_record_details and Compoid_get_community_details for metadata retrieval.
• Compoid_search_collections for searching a community’s collection tree.
• Compoid_get_collection_records for listing records within a collection.
• Record create, update, and delete operations.
• Community creation and updates with access-control and curation settings.

The resource-type vocabulary includes analysis, image, video, audio, publication, document, software, project, dataset, presentation, workflow, tutorial, and other.

Limitations and notes

Deleting a published record is irreversible and is limited to the creator or community owner. Access to files depends on their access status; the documented download capability applies to open-access records. Community and record changes can also depend on membership and permissions, and failed create or update requests may identify invalid fields or permission problems in an explanatory hint.

Compoid MCP MCP server is licensed under MIT. The remote and local deployment paths use different credential arrangements, so local configuration should not be assumed to require only the repository header used by the hosted endpoint.