MCP Tool Catalog
Every tool three.ws publishes over the Model Context Protocol, in one place:
what it does, which server hosts it, what it costs, and whether an AI client can run it without asking you first.
The build asks every server for its live tools/list, so this page cannot drift from what the servers actually serve.
Each tool has its own page with its schema, an example call and, for free read-only tools, a button that runs it.
The same data is machine-readable at /mcp-catalog.json and /tools.json.
Connect three.ws to Claude, ChatGPT, Cursor or VS Code in two clicks to use them.
How many tools, resources and prompts three.ws serves
Per-server trust grades and how these numbers are calculated
Let your assistant run the safe ones by itself
Your AI client decides whether to run a tool without asking you by trusting its read-only label. Everywhere else that label is the tool author's own word for it. Here it is checked against the code on every build, so it is worth automating on. Pick how much you want automated and paste the result into your client.
Loading the policy…Always current: /api/mcp-policy returns this same document, so a client can refetch it instead of pinning a copy that goes stale as tools are added.
Releases
Which tools each release added or removed, from /mcp-tool-history.json.
Replaying every release from the first reproduces today's catalog exactly, and the build fails when it does not.