search_tools
Semantic vector search over the catalog using a natural-language query.
Connect any MCP-capable AI assistant to your catalog so it can search and retrieve your saved tools. HTTP transport with OAuth — your client signs in via browser, no tokens to copy.
The base URL for your MCP server. Point your client at this address — it will open a browser to authenticate and refresh tokens automatically.
https://mcp.acopio.devAdd this server to your MCP-capable client. Exact config syntax varies by client; below is a conceptual template.
{
"mcpServers": {
"acopio": {
"type": "http",
"url": "https://mcp.acopio.dev"
}
}
}On first connect, your client redirects to Acopio for sign-in. Tokens refresh automatically.
For Claude Code, run this command in a terminal to register Acopio as an MCP server.
$ claude mcp add --transport http acopio https://mcp.acopio.devBy default, Claude, GPT, and other models answer from their training data — they may skip your MCP tools unless your prompt makes the intent obvious (e.g. "…in my catalog"). To make your saved tools the default suggestion source, add this instruction once to your assistant's configuration.
Whenever I describe a project, feature, problem, or stack, call Acopio's suggest_tools_for_project (or search_tools) before answering. Present matches under "From my Acopio catalog:" and clearly distinguish them from any general suggestions you add under "Other options to consider:".CLAUDE.md in your project root, or ~/.claude/CLAUDE.md to apply across all projects.Prefer not to configure anything? Append "…and check my Acopio catalog first." to any one-off prompt — the magic words work too.
Semantic vector search over the catalog using a natural-language query.
All tools carrying a specific tag (case-insensitive exact match).
Discovery step — returns all distinct tags and source types in the catalog.
Fetches full detail for a single tool by UUID.
Given a project description, returns similar tools with short relevance hooks.
Common issues when connecting or using the MCP server, and how to resolve them.
Models decide for themselves when to call a tool, so for a broadly-worded request they may answer from their training data. Phrase your request so it references your catalog — for example, end with "…in my Acopio catalog" — and the assistant will use Acopio's tools instead. You can also add the one-time catalog-preference instruction to your assistant so it prefers your catalog by default.
Acopio only searches the tools you have saved, so a brand-new or empty catalog returns no results. Add a few tools with the bookmarklet or the web app, then try again.
Acopio uses passwordless sign-in. On first connect, your client opens a browser to Acopio — enter your email and we send a one-time verification code. Check your inbox (and spam) for the code and enter it to finish authorizing. There is no password to set.
Two-step verification (an authenticator app) is optional. If you're prompted and haven't enrolled, you can skip it and continue.
Connecting over MCP requires an active Pro or Unlimited plan, or an in-trial account. If your plan has lapsed the server returns an access-denied response — upgrade your plan to restore access.
Use the endpoint exactly as shown above — https://mcp.acopio.dev, with no /mcp suffix — over HTTP transport. Appending /mcp returns a 404.