mcp-tavily
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
mcp-tavily
English | 简体中文
Web search MCP server for OpenPlayX,
backed by Tavily and running on Cloudflare Workers:
https://mcp-tavily.openplayx.com/mcp.
The app ships it as a preset external capability
(lib/services/external/external_presets.dart): on first launch it is added to
AI → 外部能力 (External capabilities), and the model sees the tool as
mcp_tavily_web_search. If the user deletes it, it is not added back.
Endpoints
Path | Description |
| MCP Streamable HTTP (JSON responses only, stateless, no |
| 405; no server-initiated stream |
|
|
Authentication: Authorization: Bearer <the app's anonymous install ID>, 16–64
characters of [A-Za-z0-9-]. It is not a secret; it only identifies an
installation for quota counting.
Tool web_search
Arguments: query (1–300 characters), count (1–10, default 8), time_range
(day / week / month / year).
Returns: {"results":[{title,url,snippet?,date?}],"cached":bool}.
Business errors are tool results with isError: true and
structuredContent {"error":{status,code,message,retryable,detail?}}, the
same shape as the other mcp-* services:
code | Meaning |
| Today's free searches are used up; |
| Search is switched off, the Tavily key is missing or invalid, or the Tavily account is out of credits |
| Tavily failed |
| Invalid arguments |
Related MCP server: groundlink-mcp
Design
The Tavily key lives only here, as the Worker secret
TAVILY_API_KEY: never in the repository, never in the app.Quotas (per UTC day):
DEVICE_DAILY_LIMITper installation,IP_DAILY_LIMITper IP andGLOBAL_DAILY_LIMIToverall. Counters live in a SQLite Durable Object (Quota), exact and atomic; upstream failures refund the quota. IPs are stored only as hashes.Cache: identical queries (case-insensitive) with the same count and time_range are served from the edge cache for 6 hours, at no quota or cost.
Kill switch: set
SEARCH_ENABLEDto"false"and redeploy.Install IDs can be changed at will, so the per-install quota is a soft limit. Cost is really bounded by the global limit plus the spending limit set in the Tavily dashboard.
When the tool definition changes, update the copy bundled in the app (
ExternalPresets.tavily.toolDefs); the app's live test compares the two.
Using it from other MCP clients
The hosted instance is meant for the OpenPlayX app and has small quotas. To use it from other clients, deploy your own (below) and configure:
{
"mcpServers": {
"tavily": {
"type": "http",
"url": "https://<your-domain>/mcp",
"headers": { "Authorization": "Bearer <any 16–64 character alphanumeric ID>" }
}
}
}Develop and deploy
Requirements: Node.js 20+; for deployment a Cloudflare account (SQLite-backed Durable Objects work on the free plan) and a Tavily API key.
npm install
npm run smoke # local end-to-end: wrangler dev + mock Tavily, small quotas
npx wrangler deploy # deploy (binds the custom domain in wrangler.jsonc routes; DNS and certificate are automatic)
npx wrangler secret put TAVILY_API_KEY # or add it in the dashboard under Settings → Variables and Secrets
npx wrangler tail # live logsFor your own deployment, change routes in wrangler.jsonc to a domain on
your Cloudflare account, or remove routes and set workers_dev to true to
use a *.workers.dev address. Variables for local development go into
.dev.vars (git-ignored).
To change quotas, edit vars in wrangler.jsonc and redeploy; the app needs
no release.
Contributing
Issues and pull requests are welcome, see CONTRIBUTING.md. Report security problems privately as described in SECURITY.md.
Disclaimer
This project is not affiliated with Tavily. Results come from Tavily and third-party websites and are not guaranteed to be accurate. If you deploy your own instance, you must follow Tavily's terms of service and you pay for its use.
License
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Live AI-native web search with citations. One tool for every MCP client. Flat per-request pricing.
Web search API and MCP server for developers building AI agents.
Web search and page fetch for AI agents. Results come from a shared cache; cached calls cost less.
Unofficial Serper MCP — wraps the Serper Google Search API (serper.dev)
Related MCP Servers
- FlicenseAqualityDmaintenanceEnables MCP-compatible agents to perform web searches via the agent-web-search engine, returning ranked results with sources.1-
- AlicenseAqualityDmaintenanceEnables MCP hosts to perform cited web searches and receive source-bearing results with titles, URLs, snippets, and sources for evidence-grounded answers.136 npmMIT
- FlicenseNot gradedqualityBmaintenanceEnables MCP clients to run Google Search, News, and Local queries and receive structured organic results, news, and local packs as JSON.-
- FlicenseNot gradedqualityBmaintenanceEnables MCP clients to run local web searches with automatic failover across multiple search providers and to extract public web pages as titles and Markdown content, using circuit breaking and key-redacted logging.-