Tavily
by Tavily
Give an agent live web search, page extraction, site mapping, site crawling and cited research. Five tools, all metered in credits — and one of them, tavily_research, can spend up to 250 credits on a single call.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect Tavily via MCP
https://mcp.tavily.com/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Tavily Tools & Capabilities (5)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- Every tool spends money. There is no free read here. Tavily meters search, extract, map, crawl and research in credits, and the connector offers no documented per-call spending cap.
tavily_researchhas a 250-credit ceiling per request. Tavily documents 15–250 credits forproand 4–110 formini. Four maximum-costprocalls exhaust the free monthly allowance.- Crawl scope is caller-controlled, not API-capped.
max_depthreaches 5 andmax_breadthreaches 500 per level.limithas a documented minimum of 1 and no documented maximum. allow_externaldefaults totrue. Results can include links beyond the domain you named unless you constrainselect_domainsorexclude_domains.- robots.txt handling is undocumented. Tavily's docs mention robots.txt twice, both times as advice to the reader. Nothing states that Tavily's own crawler reads or honours it, and no crawler user-agent is published.
- Prompt-injection filtering is undocumented. Zero mentions of injection, sanitisation or redaction across Tavily's full documentation corpus, despite the directory description advertising both PII redaction and prompt injection protection.
tavily_extractcaps at 20 URLs per call. Tavily's API reference states the list must not exceed 20.- Research is rate-limited hard. Tavily documents 20 requests per minute for research on both development and production keys, against a 1,000/minute default for production elsewhere. Crawl sits at 100/minute on both key types.
- Keyless access is Search and Extract only. Crawl, Map and Research all require an API key.
- Production keys need a paid plan. Tavily documents production rate limits as requiring an active paid plan or pay-as-you-go enabled.
- We could not verify tool behaviour ourselves. The endpoint returns 401 to an anonymous handshake, so no schemas and no safety annotations were readable. Tool names come from Anthropic's directory, corroborated by Tavily's docs; parameters come from Tavily's REST API reference.
- Prompts are unconfirmed. Anthropic's snapshot lists no prompt names, and the gated endpoint blocked our own
prompts/list, so we cannot say whether the server serves any.
Frequently asked questions
How many credits does each Tavily MCP tool cost?
It depends entirely on the tool. Tavily's pricing page prices a basic search at 1 credit and an advanced search at 2. Extract costs 1 credit per 5 successful URLs, or 2 if advanced. Map costs 1 credit per 10 pages. Crawl bills mapping plus extraction together. Research is the outlier, from 4 to 250 credits per request.
Can a single tavily_research call burn a large amount of credit?
Yes. Tavily documents dynamic pricing with per-request boundaries, and the ceiling is high: a request using the pro model consumes between 15 and 250 credits, while mini runs from 4 to 110. On the free 1,000-credit monthly plan, four maximum-cost pro calls would consume the entire month. The agent chooses, not you.
Does Tavily crawl a whole website, and does it respect robots.txt?
It crawls a site, and its documentation never states that its own crawler obeys robots.txt. Crawl defaults are depth 1, breadth 20 and a 50-link cap, but the schema allows depth 5, breadth 500 and no documented maximum on the total. Tavily's crawl best-practices page tells readers to respect robots.txt — advice to you, not a claim about itself.
Does Tavily filter prompt injection out of the web content it returns?
Its documentation makes no such claim. Anthropic's directory description advertises PII redaction and prompt injection protection, but a full-text search of Tavily's own documentation on 2026-08-19 found zero mentions of injection, sanitisation or redaction. Treat every page these tools return as untrusted text entering the model's context, because nothing documents otherwise.
Do you need OAuth to use the Tavily MCP server, or will an API key work?
Either works. Tavily documents an API key passed as a URL query parameter or an Authorization header, and OAuth 2.0 as the alternative that keeps the key out of the URL. Under OAuth, Tavily selects which key to bill by name, preferring one called `mcp_auth_default` in your personal account over the same name on a team.
Is the Tavily MCP server free to use without an account?
Partly. Tavily documents keyless access to the remote MCP server, but it covers Search and Extract only — Crawl, Map and Research all require an API key. A free key grants 1,000 credits per month with no credit card. Keyless access is rate-limited and intended for exploration rather than production traffic.
What rate limits apply to the Tavily MCP tools?
They differ per endpoint. Tavily documents 100 requests per minute on development keys and 1,000 on production keys as the default. Crawl is capped at 100 per minute on both, and Research at just 20 per minute on both. Exceeding a limit returns HTTP 429 with a `retry-after` header giving the seconds to wait.
Sources
- Tavily MCP documentation — https://docs.tavily.com/documentation/mcp (fetched 2026-08-19 as Markdown via the
.mdsuffix; this is thedocumentationURL Anthropic's directory publishes) · retrieved 2026-08-19 - Tavily Credits & Pricing — https://docs.tavily.com/documentation/api-credits (fetched 2026-08-19) · retrieved 2026-08-19
- Tavily Rate Limits — https://docs.tavily.com/documentation/rate-limits (fetched 2026-08-19) · retrieved 2026-08-19
- Tavily Crawl API reference — https://docs.tavily.com/documentation/api-reference/endpoint/crawl (fetched 2026-08-19) · retrieved 2026-08-19
- Tavily Map API reference — https://docs.tavily.com/documentation/api-reference/endpoint/map (fetched 2026-08-19) · retrieved 2026-08-19
- Tavily Research API reference — https://docs.tavily.com/documentation/api-reference/endpoint/research (fetched 2026-08-19) · retrieved 2026-08-19
- Tavily Crawl best practices — https://docs.tavily.com/documentation/best-practices/best-practices-crawl (fetched 2026-08-19; the only page on the site mentioning robots.txt, twice, both as advice to the reader) · retrieved 2026-08-19
- Tavily FAQ — https://docs.tavily.com/faq/faq (fetched 2026-08-19) · retrieved 2026-08-19
- Tavily agents setup guide — https://docs.tavily.com/agents (fetched 2026-08-19) · retrieved 2026-08-19
- Tavily docs machine index — https://docs.tavily.com/llms.txt (fetched 2026-08-19) · retrieved 2026-08-19
- Tavily full documentation corpus — https://docs.tavily.com/llms-full.txt (fetched 2026-08-19, 812 KB; searched for injection, sanitisation and redaction terms — zero matches) · retrieved 2026-08-19
docs.tavily.com/robots.txtallows the docs path and carriesContent-Signal: ai-train=yes, search=yes, ai-input=yes— synthesis is expressly permitted on all three axes (fetched 2026-08-19). The marketing hostwww.tavily.com/robots.txtallows AI crawlers by name and disallows only/api/,/legal/and one thank-you path.mcp.tavily.comserves no robots.txt (404) · retrieved 2026-08-19- Live RFC 9728 protected-resource descriptor — https://mcp.tavily.com/.well-known/oauth-protected-resource/mcp (fetched 2026-08-19; the root and path-insert forms both return 404) · retrieved 2026-08-19
- Live RFC 8414 authorization-server metadata — https://mcp.tavily.com/.well-known/oauth-authorization-server (fetched 2026-08-19) · retrieved 2026-08-19
- Live anonymous probe — 2026-08-19:
POSTto the endpoint returns HTTP 401 withwww-authenticate: Bearer resource_metadata="https://mcp.tavily.com/.well-known/oauth-protected-resource/mcp"and no scope parameter · retrieved 2026-08-19 - Anthropic connector directory — https://claude.ai/directory/d8a25d2a-e5ea-4860-b990-277244df5417 (snapshot 2026-08-16) · retrieved 2026-08-16
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Data & Research
- Developer
- Tavily
- Tools
- 5
- Domain
- mcp.tavily.com
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.