Agent interface

Context Factory MCP

Connect an MCP-compatible host to search the public catalog. Same metadata-only discovery contract as the API.

PublicRead-onlyNo authStreamable HTTP

Endpoint

Streamable HTTP endpoint
https://mcp.contextfactory.dev/mcp

Public, read-only, and stateless. No account, API key, or OAuth is required to connect.

Add the server

Claude Code:

Terminal
claude mcp add --transport http context-factory https://mcp.contextfactory.dev/mcp

Cursor, VS Code, Claude Desktop, or any client that reads an mcp.json:

mcp.json
{
  "mcpServers": {
    "context-factory": {
      "type": "http",
      "url": "https://mcp.contextfactory.dev/mcp"
    }
  }
}

Add to Cursor in one click.

Tools

search_artifacts
  q?: string
  type?: markdown | text | agents | skill | readme
  owner?: string
  repository?: owner/repository
  cursor?: string
  limit?: 1-50

get_artifact
  artifactId: 12-character Context Factory ID

get_collection
  collectionId: 12-character Context Factory ID
  q?, type?, owner?, repository?, cursor?, limit?

Each tool returns validated structured content plus a JSON text fallback. Results include canonical Context Factory URLs and GitHub directory, file, and raw URLs. They never include source bodies.

Typical flow

  1. Use search_artifacts when the user gives a topic, name, owner, or repository.
  2. Use get_artifact for a known artifact URL or ID.
  3. Use get_collection for a known collection URL or ID.
  4. Let the consuming agent decide whether to open any returned source URL.

Verify

Terminal
npx @modelcontextprotocol/inspector "https://mcp.contextfactory.dev/mcp"

Validation failures come back as actionable tool errors. Start a fresh search when a cursor is invalid or expired.