The EffectHub MCP server

Every client below only needs the endpoint URL. Menu names move between releases; if yours differ, look for “custom connector” or “MCP server”, and choose no authentication.

claude mcp add --transport http effecthub https://effecthub.net/api/mcp

{
  "mcpServers": {
    "effecthub": {
      "url": "https://effecthub.net/api/mcp"
    }
  }
}

{
  "servers": {
    "effecthub": {
      "type": "http",
      "url": "https://effecthub.net/api/mcp"
    }
  }
}

Four, all read-only. Each answer is readable text for the model plus structured data for the client, and each one links the card it came from.

• search_toolsSearch AI tools

  Search the EffectHub catalog of AI tools by product name, job to be done ("transcribe interviews") or category ("video editing"). Optionally narrow by task slug, category slug or pricing model. Returns published, live tools with pricing, Effect score and a link to each card on effecthub.net.

  • query · required — What to look for: a product name, a job, or a category in plain words.
  • task · optional — Optional task slug, as in https://effecthub.net/ai-for/{task} (e.g. "text-to-speech"). Only tools tagged with it are returned.
  • category · optional — Optional category slug, as in https://effecthub.net/ai-tools/{slug}. A top-level category includes its subcategories; matched on each tool's primary category.
  • pricing · optional — Optional pricing model: free, freemium, free_trial, paid or contact (sales-quoted).
  • limit · optional — How many tools to return (1-20, default 10).

• get_toolGet an AI tool's card

  Everything public on one EffectHub tool card: what it does, category, pricing model and entry price, Effect score, Trust score, platforms, whether it is still alive and when that was last checked, and the card URL. Takes the tool's slug (from search_tools, or the last part of https://effecthub.net/ai/{slug}).

  • slug · required — The tool's slug, e.g. "cursor".

• list_alternativesList alternatives to an AI tool

  Tools that do the same job as the given one. Alternatives our research named for this tool come first (basis "researched"), then live tools from the same category (basis "category"). Discontinued tools are left out.

  • slug · required — The tool to find alternatives for.
  • limit · optional — How many tools to return (1-20, default 10).

• compare_toolsCompare two AI tools

  Two EffectHub tool cards side by side — pricing, scores, platforms, API, liveness — plus the differences in plain sentences. When EffectHub has a researched head-to-head page for the pair, comparison_url links to it.

  • a · required — Slug of the first tool.
  • b · required — Slug of the second tool.

The Effect and Trust scores are explained on how we verify. Each comes with how much it rests on — effect_signals of 4, trust_checks_measured of 8 — and is null when nothing behind it has been measured yet.

The handshake an older client opens with, then a search:

curl -s https://effecthub.net/api/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-11-25","capabilities":{},"clientInfo":{"name":"curl","version":"1"}}}'

curl -s https://effecthub.net/api/mcp \
  -H 'Content-Type: application/json' \
  -H 'MCP-Protocol-Version: 2025-11-25' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"search_tools","arguments":{"query":"transcription","pricing":"free","limit":3}}}'

• 120 requests a minute per network address; past that the server answers 429 with a Retry-After header.
• Up to 20 tools per answer, queries up to 128 characters, request bodies up to 64 KB.
• A tool that has not answered in 8 seconds returns an error instead of hanging.
• One JSON-RPC message per request; batches are refused.

• Free. No account, no key, no cookie.
• Attribution. When you show our data, link the card it came from — every result carries its URL — and credit it as “Data: EffectHub (effecthub.net)”.
• No guarantees. The data is what the catalog holds, as is: a price can change between our checks, scores are our method rather than a verdict, and the service and its limits may change.
• Only what is public. The server returns published cards and nothing else — never submissions, unpublished listings, accounts, email addresses or internal ids. We log which tool ran and whether it worked, not the query and not the address it came from.

Something wrong on a card? Tell us through contact.