brave-search-mcp-server
io.github.brave/brave-search-mcp-server
About brave-search-mcp-server
brave-search-mcp-server is an MCP server published by the GitHub account brave. Brave Search MCP Server: web results, images, videos, rich results, AI summaries, and more.
It runs on your own machine: your AI client starts it from the npm package @brave/brave-search-mcp-server at version 2.1.3.
It asks for one secret: BRAVE_API_KEY. Give it a key with the least access that still works.
Version 2.1.3 was published on 20 August 2026, and the source code is public, so you can read it before installing.
Written from the registry entry. 9 automated checks on the registry entry and the package code: 6 pass, 1 warn, 0 fail, 2 info. How we check.
What it asks for
| Variable | Required | Secret | Description |
|---|---|---|---|
BRAVE_API_KEY | Yes | Yes | Your API key for the service |
Checks
| Check | Result | Detail |
|---|---|---|
| Source repository linked | Pass | https://github.com/brave/brave-search-mcp-server |
| Version pinned | Pass | Every package has an explicit version |
| Secrets requested | Info | Asks for 1 secret: BRAVE_API_KEY |
| Known package registry | Pass | All packages come from npm, PyPI, OCI, NuGet or MCPB |
| Namespace ownership | Info | Name verified via GitHub account brave |
| Registry text aimed at the AI | Pass | Description and settings contain no hidden characters or instructions to the AI |
| Install scripts | Warn | Runs at install: prepare: npm run format && npm run build |
| Code red flags | Pass | Read 57 files in @brave/brave-search-mcp-server@2.1.3: no obfuscation, encoded execution, credential reads or data-collection endpoints found |
| Text aimed at the AI | Pass | No hidden characters or instructions to the AI in 57 files, including comments and the README |
Code review: Pass
We read all 3,717 lines in 55 files of @brave/brave-search-mcp-server 2.1.3 on 2026-10-31, without running it.
Brave Search lets the AI search the web, news, images, videos and places. It only talks to Brave's own search API, and only when a tool is used, sending the search text and your Brave API key. It reads no files besides an optional key file and a .env, runs no commands and has no telemetry.
What we found
- Info All outbound traffic goes to a single hard-coded host, api.search.brave.com. Goggle URLs are only accepted if https and are forwarded to Brave, not fetched locally (index.js:21-40).
dist/BraveAPI/index.js:45,104 - Info Optional HTTP transport binds 127.0.0.1 by default and has a DNS-rebinding Origin/Host guard. It has no client authentication, so binding it to 0.0.0.0 would expose the user's API key quota.
dist/protocols/http.js:82; dist/config.js:20,44; dist/protocols/rebinding.js:96-121 - Info Location fields (lat/long, city, postal code) are sent to Brave only when the model fills them in.
dist/tools/llm_context/schemas/input.js:178-224; dist/tools/place_search/schemas/input.js:159-170 - Info dotenv reads .env from the current working directory at startup.
dist/config.js:6 - Info Test files are shipped in the package (55 files incl. tests). They are inert unless run with node --test; reviewed and harmless.
dist/*.test.js, dist/testUtils/mcpTestHarness.js - Info HTTP session map is never pruned and registerSigIntHandler is unused; minor resource quality issue in HTTP mode only.
dist/protocols/http.js:15,41-46; dist/helpers.js:1-10 - Info The manifest file was not in the reviewed tree; dependencies listed are inferred from import statements. None look unusual or typosquatted for the job.
package.json/pyproject
| Talks to | api.search.brave.com: search query and tool parameters as query string; X-Subscription-Token API key; optional location / user-agent / api-version headers when the model supplies them (llm_context, place_search) (only when a tool is called) |
|---|---|
| Keys and settings | BRAVE_API_KEY (or --brave-api-key): api.search.brave.com as X-Subscription-Token headerBRAVE_API_KEY_FILE (or --brave-api-key-file): local file read to obtain the key, then as aboveBRAVE_MCP_LOG_LEVEL, BRAVE_MCP_TRANSPORT, BRAVE_MCP_ENABLED_TOOLS, BRAVE_MCP_DISABLED_TOOLS, BRAVE_MCP_PORT, BRAVE_MCP_HOST, BRAVE_MCP_ALLOWED_ORIGINS, BRAVE_MCP_ALLOWED_HOSTS, BRAVE_MCP_STATELESS: local configuration only |
| Files on your machine | dotenv loads a .env file from the working directory (config.js:6). Reads the API key file the user names (utils.js:23-35). Shipped test files write throwaway temp files under the OS temp dir, but only when run under node:test; they are not loaded by the server. Nothing reachable from tool input touches the file system. |
| Runs programs | None. |
| Telemetry | None. |
| Tools | brave_web_search, brave_local_search, brave_video_search, brave_image_search, brave_news_search, brave_summarizer, brave_llm_context, brave_place_search |
| Tool descriptions | clean. No hidden or invisible text. brave_local_search tells the model to confirm the user's plan before using it (tools/local/index.js:11); news/llm_context descriptions ask the model to cite sources. Ordinary guidance. |
This covers the package's own code only, not the 6 packages it depends on (@modelcontextprotocol/sdk, commander, dotenv, express, zod, ajv (tests only)). A new version drops this review until we read it again.
Popularity & health: 79 / 100
Worked out from public numbers and our own checks, not from user reviews or ratings. It says how used and how looked-after a server is, not whether it is safe.
224 forks, 255 merged pull requests. Sources: GitHub, npm, PyPI (pypistats.org) and Hacker News (Algolia), fetched 2026-10-31.
What people say
No Hacker News threads link to its source code.
Look further: search Reddit, search Hacker News, read its GitHub issues. These are not reviews we checked.
Facts
| Registry name | io.github.brave/brave-search-mcp-server |
|---|---|
| Publisher | GitHub account brave |
| How the name is verified | Name verified via GitHub account brave. The registry ties the name prefix to that account or domain; it does not check the code. |
| Type | Local: runs on your machine from a package |
| Packages | npm @brave/brave-search-mcp-server 2.1.3 (stdio) |
| Version | 2.1.3 |
| Published | |
| Last updated | |
| Source repository | https://github.com/brave/brave-search-mcp-server |
| Website | None linked |
| Category | AI & media generation (assigned by keyword rule, not by hand) |
Related servers in AI & media generation
- YouTube Translate3/3Transcripts, translations, summaries, subtitles, and search for…
- yt-mem-ai7/7Local YouTube memory: transcribe, embed and search videos, with…
Agent Data Tools: AI4/425 tools: LLM chat, summaries, translation, embeddings, speech,…- source-library7/7Search 22,000+ rare pre-modern texts with AI English translations,…
- VIDIA8/8VIDIA AI video production: get quotes, start videos, track progress…
Google Videos (via Apify)3/3Google video results for AI agents: YouTube, TikTok and more, with…