Index Agentica

What Index Agentica MCP server does

Index Agentica MCP server exposes the published Index Agentica catalog to MCP clients and HTTP consumers. The catalog covers agent skills, harnesses, MCP servers, tools, protocols, APIs, guides, comparisons, stacks, and downloadable skills. Access is read-only: the service searches and retrieves published data but does not modify the source catalog.

The remote MCP endpoint is available at https://mcp.indexagentica.com/mcp. A separate REST interface provides search, entry lookup, content retrieval, category listing, OpenAPI documentation, health checks, and public usage statistics. No API key or authentication is required.

How it works

The service runs as a stateless Cloudflare Worker. MCP requests use Streamable HTTP and return JSON rather than an SSE stream. Each request is independent, and the server does not issue MCP session IDs. It supports the current protocol behavior described in the project documentation as well as several legacy protocol versions, including initialization, tool listing, tool calls, ping, and legacy batch requests.

Catalog data comes from Index Agentica’s published static APIs. Entry and category indexes are fetched together, while long-form content is fetched when requested. The Worker keeps indexes in isolate memory and Cloudflare Cache API storage for the configured cache period. If an upstream refresh fails, an available stale copy can still be served.

Setup and configuration

For a remote client, add https://mcp.indexagentica.com/mcp as a Streamable HTTP MCP server with no authentication. Claude Code, for example, can use:

claude mcp add --transport http indexagentica https://mcp.indexagentica.com/mcp

The Worker also exposes https://indexagentica-mcp.indexagentica.workers.dev/mcp. CORS is open by default because the catalog is public, though deployments can restrict origins with ALLOWED_ORIGINS. The implementation also supports configuration for the upstream data base URL and cache duration; these are deployment concerns rather than requirements for using the hosted endpoint.

Tools and capabilities

Index Agentica MCP server provides four read-only tools:

• search ranks matching entries and long-form items. It accepts a query, content type, entry category, tags, and result limit. Multiple tags are treated as an AND filter.
• get_entry returns a complete entry by kebab-case ID, including links and references to related long-form material.
• get_content retrieves a guide, comparison, stack, or skill. It can include HTML; skills also expose the full skill document and downloadable links.
• list_categories returns category names, descriptions, counts, and links.

Tool responses include structured JSON and an equivalent text content block. Input and output schemas are exposed, and read-only annotations identify the tools as non-mutating. Invalid arguments and unknown IDs produce tool errors that an MCP client can handle.

Limitations and notes

The service is a directory lookup layer, not an execution environment for the listed tools or servers. Results depend on the published Index Agentica data and its upstream availability. The REST /mcp endpoint accepts POST for MCP requests; GET and DELETE return method-not-allowed responses because there is no standalone SSE stream or session lifecycle.

The hosted service has no authentication and exposes public catalog data. Deployments can configure allowed origins. Usage events may be recorded in Cloudflare Workers Analytics Engine when the deployment has the relevant binding, but logging failures do not affect responses.