MCP
MCP
The Pureframe MCP server exposes your video library as a set of tools any MCP-compatible AI agent can call. Once connected, agents can search footage, list collections, and retrieve video details — all through natural language.
Available tools
Local MCP server
Install via npx — no global install needed:
Add this config to:
~/Library/Application Support/Claude/claude_desktop_config.json
~/.claude.json
.cursor/mcp.json
~/.codeium/windsurf/mcp_config.json
Restart your client after adding the config.
Remote MCP (no install)
For clients that support remote MCP endpoints — including Claude.ai web:
No npx or local install required. The server runs on Pureframe’s infrastructure.
What agents can see
search_videos results include thumbnail_base64 — a base64-encoded JPEG of the matched frame. Vision-capable models can see this image directly without fetching a URL, enabling follow-up questions about visual content:
“Find clips where someone is pointing at a whiteboard, then describe what’s written on it.”
Rate limits
Agent tool calls share the same rate limits as the REST API — 30 requests per minute for search.
Scoping to a collection
Pass a collection_id to search_videos to limit results to a specific library:
Use list_collections first to discover available collection IDs.