Agent Search MCP
Agent Search MCP searches in English and Chinese with no API key, returns multi-source evidence and records provider failures.
- 113 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Search queries are sent to external search engines.
- Paid providers spend money when their keys are set.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/agent-search-mcpDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Search queries are sent to external search engines.
- Paid providers spend money when their keys are set.
This is third-party code. Review the repository files before installing.
What it does
The Node.js server starts without a key and queries nine free engines: DuckDuckGo, Bing, Startpage, Yandex, Mojeek, Wiby, Wikipedia and, for Chinese queries, Sogou and Baidu. A result carries links, relevance and the count of independent sources, counting provider families rather than adapters. A provider failure, timeout or challenge is not hidden behind an empty reply: it stays in `partialFailures`. Paid Brave, Tavily, Exa, Serper and others join only by key and routing policy. Budgets for calls, time and evidence characters bound the work and the reply.
Who it is for. Developers and researchers who want agent web search without a provider account, including Chinese-language sources.
Good fit when
- An agent verifies a claim against several independent sources.
- A Chinese query goes through Sogou and Baidu with no translation.
- A full page behind a link is read as clean Markdown.
- Paid search is allowed only when free quality falls short.
- A GitHub repository README is fetched in one call.
Not a fit when
- You need a stable commercial SLA: free engines may return a challenge or a rate limit.
- Queries must not leave for external search engines.
- A recency filter is mandatory: `time_range` currently returns `UNSUPPORTED_FILTER`.
Example request
Verify this claim against official sources and show how many independent sources support it.MCP
- Transport
- stdio, http
- Authentication
- not required
Environment variables
SEARCH_PROVIDER_MODE- Routing mode: `free_first`, `quality_escalation`, `paid_first` or `free_only`
BRAVE_API_KEYsecret- Optional key for paid Brave search; TAVILY_API_KEY, EXA_API_KEY and others are set the same way
DISABLED_TOOLS- Tools to disable; deny wins over allow
Limitations
Node.js 18.17 or newer is required. Free adapters rely on public search pages, so availability and quality shift; the server surfaces such failures and does not remove them. Paid providers run on your own keys, and trial quotas are not treated as permanent. Queries go to the selected search engines. Benchmark quality scores in the repository do not replace a test on your own queries.
How to disable. Remove the `agent-search` entry from the client configuration. If you added the Skill, delete its folder; for a global install run `npm uninstall -g agent-search-mcp`.
FAQ
Is a key required?
No. Nine adapters work keyless. Brave, Tavily, Exa, Keenable, You.com, Tencent, Bocha and Serper keys are optional extras.
Which tools are there?
`free_search`, `free_search_advanced`, `free_extract`, `fetch_github_readme`, `fetch_csdn_article`, `fetch_juejin_article` and `search_with_synthesis`. All are read-only.
How do I block paid providers?
Set `SEARCH_PROVIDER_MODE=free_only`. The default is `free_first`, and paid search runs only by policy and with a key.
How do I check the install?
Run `fasm doctor` after a global install: it checks the local runtime without a search request.
How do I read pages behind links?
Call `free_extract` with a page address and it returns clean Markdown. Separate tools handle GitHub READMEs and CSDN and Juejin articles.
Discussion
No questions or reviews yet.
Related
MCP serversBrave's official MCP server: web, news, image, video and place search plus extracted page content for model context
A unified search MCP server: one interface to Tavily, Brave, Kagi, Exa, Linkup, Firecrawl and GitHub search
Web search via DuckDuckGo and page reading for the agent with no API keys, with built-in rate limiting
Metasearch for agents: one MCP server and HTTP API merge results from many search providers into unified JSON with deduplication.
MCP server for real-time web search: general and vertical search, batch queries and page content extraction
A web search MCP server that gives the agent full page text at once: Stack Overflow answers, GitHub threads, arXiv papers
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Agent Search MCP (this card) | no | not required | 113 | |
| free-search-mcp | no | not required | 179 | also searches without a key; Agent Search MCP reports independent source counts, provider failures and supports Chinese engines |
| Brave Search MCP Server | yes | API key | 1,577 | runs on a single paid Brave API; here nine free adapters come with optional paid ones |
README in short
The server runs over stdio and Streamable HTTP with a `fasm` CLI. Budgets cap calls, time and evidence size, and an optional Skill picks one of four modes: quick discovery, verification, Chinese sources, page extraction.