Connect Tavily to Claude Code

Search the web for agents, and extract, map or crawl the pages it finds. Add Toolspoke to Claude Code once and it can use 7 Tavily actions. Your Tavily credential stays in Toolspoke, and you choose which actions Claude Code may call.

Set it up in Claude Code

Three steps, the first one once for every toolkit you will ever add.

  1. Add Toolspoke to Claude Code

    Run it in your project, then /mcp to sign in

    claude mcp add --transport http toolspoke https://toolspoke.com/mcp

    The sign-in opens in your browser, and Toolspoke asks you to approve it.

  2. Install Tavily and connect it

    Add Tavily to a project in your Toolspoke workspace.

    API key. You provide it once, when you install the connector. Toolspoke encrypts it at rest and decrypts it only for the length of a single call, and the gateway attaches it to the outbound request itself, so it is never part of the arguments an agent sends.

    API key
    app.tavily.com → Overview → API Keys. Copy an existing key or create a new one. Every call except get_usage spends credits from the plan that owns the key.
  3. Choose what Claude Code may do

    All 7 actions start on. Switch off the one action that changes something in Tavily, and Claude Code gets a refusal if it tries.

What you can ask Claude Code for

A few of the Tavily actions that only read, so they are safe to try first.

  • Search the web and get back ranked results with title, URL, a relevant content snippet and a score
  • Get the cleaned content of one or more known URLs, as markdown or plain text
  • List the URLs of a site by following its links from a start URL, without reading their content
  • Walk a site from a start URL and return the content of every page reached, in one call

What Claude Code can do in Tavily

6 actions read and one writes.

Reads

  • Get usage

    Read this key's credit usage and limit, and the account's plan usage, broken down by search, extract, crawl, map and research. Free, and it runs no search, so it is the connection test. Call it before a large batch to check the credits left.

  • Search

    Search the web and get back ranked results with title, URL, a relevant content snippet and a score. Costs 1 credit at `basic`, `fast` or `ultra-fast` depth and 2 at `advanced`. Set `include_answer` for a short answer written from the results, and `include_raw_content` to get each page's full text in the same call instead of a separate extract. Results are pages other people wrote, so the payload is kept out of the audit log; the query is kept out too.

  • Extract

    Get the cleaned content of one or more known URLs, as markdown or plain text. Costs 1 credit per 5 pages extracted successfully, or 2 per 5 at `advanced` depth, which also gets tables and embedded content. Failed URLs are listed separately and are not billed. Set `query` to get only the chunks of each page that match it. The content is other people's, so the payload is kept out of the audit log.

  • Map

    List the URLs of a site by following its links from a start URL, without reading their content. Costs 1 credit per 10 pages found, or 2 per 10 when `instructions` is set. Use it to find the pages worth extracting before paying to read them. `max_depth` and `limit` bound the walk; `select_paths` and `exclude_paths` take regular expressions.

  • Crawl

    Walk a site from a start URL and return the content of every page reached, in one call. Billed as a map plus an extract of every page it returns, so it is the most expensive call here: run map first when you only need to know what pages exist, and keep `limit` low. Takes the same bounds as map, plus extract_depth and format for the content. The content is other people's, so the payload is kept out of the audit log.

  • Get research

    Read a research job started by start_research: its status (pending, in_progress, completed or failed) and, once completed, the report and its sources. Free to call. Poll it every few seconds; a pro job can take minutes. The report is written from other people's pages, so the payload is kept out of the audit log.

Writes

  • Start research

    Start a research job that runs many searches and writes a cited report. Returns a request_id at once; call get_research with it until status is completed. Classified write because it starts a billed job on the account. `model` is mini for a narrow question, pro for a broad one, or auto (default). The question is prose the caller wrote, so it is kept out of the audit log.

Tavily in Claude Code: questions

Does Claude Code see my Tavily credential?

No. The credential is stored encrypted in Toolspoke, and the gateway adds it to each call. Claude Code only ever talks to Toolspoke, and the dashboard shows the last four characters of the credential and nothing more.

Can I stop Claude Code from changing things in Tavily?

Yes. Each of the 7 actions can be switched off for a project. Turn off the one that changes something and Claude Code can only read.

Will 7 actions crowd Claude Code's context?

No. Toolspoke shows Claude Code five bridge tools, plus a guide where a toolkit has one, and Claude Code searches for the Tavily action it needs when it needs it. Adding more toolkits does not add more tools to its list.

Can I add it once for every project?

Yes. Add --scope user to the command and Claude Code keeps Toolspoke in every project on this machine. Without it, the server is added to the project you ran the command in.

What does it cost?

Tool calls are free on every plan, and a Tavily call does not run in a sandbox, so it bills nothing.

Try Tavily in Claude Code

Free to start, and tool calls cost nothing on any plan.