Interfaces

For agents

The catalogue is not just for reading, you can also query it. Agents like Claude or Cursor find apps, libraries, skills and MCP servers here, and get how to set them up right along with them.

No account
No key
Read-only
Four languages

MCP server

The easiest way: add Collected.tools as a connector. Your agent then asks on its own in the middle of a task, for example “Is there an MCP server for Figma?”.

Address

https://collected.tools/mcp

Claude

Open Settings in Claude and go to “Connectors”.
Choose “Add custom connector”.
Enter Collected.tools as the name and the address above as the URL.

Claude Code

One command in the terminal:

claude mcp add --transport http collected-tools https://collected.tools/mcp

Cursor

In .cursor/mcp.json in your project, or ~/.cursor/mcp.json for all projects:

{
  "mcpServers": {
    "collected-tools": {
      "url": "https://collected.tools/mcp"
    }
  }
}

Other clients

Clients that only know local servers over stdio connect through mcp-remote:

{
  "mcpServers": {
    "collected-tools": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://collected.tools/mcp"
      ]
    }
  }
}

Tools

search_tools: Searches the catalogue, with filters for type, category, pricing and open source. Returns summaries.
get_tool: One entry with every detail, including repository and install instructions.
list_categories: All types and categories with descriptions and entry counts.

JSON API

For your own scripts, automations and anything that does not speak MCP. The same filters as the directory, with the same names. Type, category and price take several comma-separated values, and an entry only has to match one of them.

https://collected.tools/api/v1

Endpoints

GET /tools: Search and filter. Returns summaries, full entries with full=1.
GET /tools/{slug}: One entry with all fields, including install instructions.
GET /categories: All categories, that is, what a tool is for.
GET /types: All types, that is, how a tool is used.
GET /openapi.json: Description of the API in OpenAPI format.

Filters

q: Search words. Every word must appear in the name, description, type or tags.
type: app, library, agent, skill, mcp, prompt, template or api.
category: design, assets, animation, frontend, backend, hosting, devtools, ai, nocode, automation, marketing, analytics, video or productivity.
price: free, freemium, paid or trial.
oss: 1 for open source only.
lang: Language of the texts: de, en, uk or pl. Defaults to en.
limit, offset: Page through results, 100 per request at most.
full: 1 for full entries instead of summaries.

Example: libraries for frontend or animation

curl "https://collected.tools/api/v1/tools?type=library&category=frontend,animation"

Open the OpenAPI description

llms.txt

For agents that research on the web: the whole catalogue as Markdown, without HTML around it. Following the llmstxt.org standard.

/llms.txt: Short version with one sentence per entry.
/llms-full.txt: Everything on the detail pages, including install instructions.

Ground rules

Everything is free to fetch and read-only. Submissions still go through the form.
At most 120 requests per minute and IP. Above that, the IP is blocked for five minutes.
Responses are cached for up to five minutes, so new entries show up with a short delay.
The shape of responses only changes with a new version under /api/v2. What is under /api/v1 stays.