AgentSidekick

MCP server

AgentSidekick has a hosted Model Context Protocol server, so an AI agent can search this directory and read product records directly. It exposes 7 read-only tools over the 9,469 profiled products, the platform records, market segments and scored opportunities. It needs no account or key.

Connect

Endpoint (Streamable HTTP):

https://mcp.agentsidekick.dev/mcp

Claude Code

claude mcp add --transport http agentsidekick https://mcp.agentsidekick.dev/mcp

Cursor and other clients that read an mcp.json file

{
  "mcpServers": {
    "agentsidekick": {
      "url": "https://mcp.agentsidekick.dev/mcp"
    }
  }
}

Claude, ChatGPT and other apps with custom connectors: add a custom connector or remote MCP server and paste the endpoint URL. Menu names differ between apps and change often, so check the app's own instructions.

The client commands above follow each vendor's documented format as of 2026-11-21; if one has changed, the endpoint URL is the only thing you need.

Tools

All tools only read data. They return the same records this site shows, with the same evidence labels.

ToolWhat it returnsArguments
search_productsSearch profiled products by keywords and filters. Ranking: keyword match (name, then capability/integration tags, then description), ties broken by GitHub stars of the product's own repository. Returns up to `limit` results with URL, type, publisher, licence, tags and vendor-claimed platforms.query?, type?, capability?, integration?, platform?, license?, limit?
get_productFull record for one product: description, licence, hosting, install packages and endpoints, capabilities, integrations, per-platform compatibility status with the reason, permission signals, known advisories, evaluation dimensions, closest alternatives, and the provenance of each field.slug
compare_productsSide-by-side facts for 2 to 4 products: type, licence, hosting, stars, last push, capabilities and integrations shared by all versus unique to each, and compatibility statuses. Unknown values are returned as null, never assumed.slugs
list_platformsAgent platforms (Claude Code, Cursor, Codex, ChatGPT, Gemini CLI…) with vendor, kind, whether each documents MCP and Agent Skills support, extension mechanisms and lifecycle. Researched records with sources and a check date.query?
market_segmentsCategories, capabilities and integrations with the number of indexed products and a momentum label. Use it to find slugs for search_products filters and to see which areas are crowded or thin. Counts cover the whole catalogue.query?, kind?, limit?
list_opportunitiesGaps in the ecosystem scored on six weighted components (demand, competition gap, monetization, growth, feasibility, distribution), each with evidence for and against marked measured or heuristic. "integration-gap" has measured demand; "underserved-capability" is a hypothesis.kind?, query?, limit?
recent_changesMost recent releases, field changes and removals recorded for profiled products.type?, slug?, limit?

What to know before relying on it

  • Search covers profiled products only: 9,469 of 73,820 indexed. The rest are counted in segment totals but cannot be searched or fetched.
  • Compatibility is a status, not a test result. "vendor-claimed" means the publisher names the platform; "protocol" means the platform documents the standard. No pairing has been independently verified.
  • Tags come from keyword rules and can be wrong or missing. Each product record shows the keyword that matched.
  • Signals and advisories are not a safety verdict. None found is not evidence of safety.
  • Data is as fresh as the last site build (2026-11-21); the server caches responses for five minutes.
  • Rate limit: about 120 requests a minute per client. There is no uptime guarantee.

The same data is available as plain JSON without MCP: see the dataset links in the methodology.

Page data last updated .