Agents are good at using tools and bad at knowing which tools exist. Ask a coding agent to “find an AI tool that transcribes interviews and exports SRT” and it will guess from training data that may be a year old. Connecting it to a live directory fixes that: the agent searches, reads the listings and cites them.
What the EveryAI MCP server does
The server exposes the reviewed EveryAI index over the Model Context Protocol. Every tool but one only reads, so connecting it cannot change anything on your machine. The exception, submit_tool, lists a founder’s tool on EveryAI, and nothing is published until the founder confirms by email or pays (see For AI agents).
| Tool | What it returns |
|---|---|
| search_tools | The best-matching tools for a task (“logo maker”, “transcribe podcast episodes”), with EveryAI page, website, category and a short description. Optional category filter and limit. |
| get_tool | Full public details of one tool by its EveryAI URL, slug, website or exact name. |
| list_categories | Every category and use case, with hub pages and tool counts. |
| check_listing | Whether a website already has a live listing. |
| get_featured_offer | What a Featured listing costs in the founder’s country. |
| prepare_submission | A prefilled submit link for a founder. Nothing is submitted automatically. |
Results marked featured are paid placements, and the server says so in every response, so an agent can tell the difference.
Connect it to Claude Code
Run this once in a terminal:
claude mcp add --transport http everyai https://api.every-ai.com/mcp
Then ask in plain language: “Use everyai to find three tools for turning meeting recordings into action items.” The agent calls search_tools, reads the matches and can open any of them with get_tool.
Connect it to other MCP clients
Any client that supports remote (Streamable HTTP) servers takes the same URL. Clients that use an mcpServers JSON file, such as Cursor, accept an entry like this:
{
"mcpServers": {
"everyai": { "url": "https://api.every-ai.com/mcp" }
}
}
Check your client’s documentation for where that file lives. The discovery card at /.well-known/mcp.json describes the server for clients and registries that read it.
Use the HTTP API instead
Scripts and agents without MCP can call the same search over HTTPS:
curl "https://api.every-ai.com/api/tools/search?q=transcribe%20interviews&limit=5"
Results come back best match first as JSON. /api/tools/check?url=https://example.com tells you whether a site is listed, and the full contract is in the OpenAPI document. Please keep to about 120 requests a minute.
Prompts that work well
- “Find AI tools for [task] and compare the top three on what they do and who they are for. Link each EveryAI page.”
- “List EveryAI’s categories and tell me which use cases have the most tools.”
- “Is [website] listed on EveryAI? If not, list it for me with my email.”
Skills, too
EveryAI also catalogs agent skills. The open-source skills MCP gateway in the EveryAI repository searches that catalog and reads a skill’s SKILL.md as untrusted text without installing it. See the MCP guide and the skills library.