Capacium MCP Server

Note

Public mirror. The canonical repository is hosted on our self-hosted git. This GitHub copy is a read-only mirror kept in sync for visibility and installation. Bug reports are welcome via Issues; pull requests are applied upstream and synced back here.

Model Context Protocol (MCP) server exposing Capacium Exchange capabilities to AI agents.

Resources

• capacium://system/status — System health and stats
• capacium://{type}/{query} — Search capabilities (e.g., capacium://search/claude)
• capacium://popular — Popular capabilities
• capacium://capabilities/{owner}/{name} — Detail view
• capacium://crawler/sources / capacium://crawler/status — Crawler management

Tools

• cap_install — Install a capability
• cap_verify — Verify fingerprint
• cap_status — Check install status
• crawler_source_enable / crawler_source_disable — Control sources
• crawler_doctor — Health check
• crawler_cycle — Trigger crawl cycle

Usage

pip install capacium-crawler
cap mcp start                  # stdio transport (default)
cap mcp start --transport sse --port 9999

Agents connect via MCP protocol to browse, search, install, and verify capabilities.

Development

Canonical repository: self-hosted Forgejo — git.langevc.com/capacium/capacium-mcp (git clone git@git.langevc.com:capacium/capacium-mcp.git). Develop against the Forgejo clone and open pull requests there. The GitHub copy is a read-only mirror.