Skip to main content
Glama
enesjakozh
mikechao

Brave Search MCP

by mikechao
OverviewSchemaSourceRelated ServersDiscussionsAdmin
TypeScript
Remote

Brave Search MCP Server

An MCP Server implementation that integrates the Brave Search API, providing, Web Search, Local Points of Interest Search, Video Search, Image Search, News Search and LLM Context Search capabilities

Features

  • Web Search: Perform a regular search on the web

  • Image Search: Search the web for images.

  • News Search: Search the web for news

  • Video Search: Search the web for videos

  • Local Points of Interest Search: Search for local physical locations, businesses, restaurants, services, etc

  • LLM Context Search: Fetch and extract full web page content optimized for reading and synthesizing sources

Related MCP server: Brave Search With Proxy

Tools

  • brave_web_search

    • Execute web searches using Brave's API

    • Inputs:

      • query (string): The term to search the internet for

      • count (number, optional): The number of results to return (max 20, default 10)

      • offset (number, optional, default 0): The offset for pagination

      • freshness (enum, optional): Filters search results by when they were discovered

        • The following values are supported

          • pd: Discovered within the last 24 hours.

          • pw: Discovered within the last 7 Days.

          • pm: Discovered within the last 31 Days.

          • py: Discovered within the last 365 Days

          • YYYY-MM-DDtoYYYY-MM-DD: Custom date range (e.g., 2022-04-01to2022-07-30)

  • brave_image_search

    • Get images from the web relevant to the query

    • Inputs:

      • query (string): The term to search the internet for images of

      • count (number, optional): The number of images to return (max 50, default 10)

  • brave_news_search

    • Searches the web for news

    • Inputs:

      • query (string): The term to search the internet for news articles, trending topics, or recent events

      • count (number, optional): The number of results to return (max 20, default 10)

      • offset (number, optional, default 0): The zero-based offset for pagination (max 9)

      • freshness (enum, optional): Filters search results by when they were discovered

        • The following values are supported

          • pd: Discovered within the last 24 hours.

          • pw: Discovered within the last 7 Days.

          • pm: Discovered within the last 31 Days.

          • py: Discovered within the last 365 Days

          • YYYY-MM-DDtoYYYY-MM-DD: Custom date range (e.g., 2022-04-01to2022-07-30)

  • brave_local_search

    • Search for local businesses, services and points of interest

    • Falls back to brave_web_search on the initial page if no location results are found

    • Inputs:

      • query (string): Local search term

      • count (number, optional): The number of results to return (max 20, default 5)

      • offset (number, optional, default 0): The zero-based offset for pagination (max 9)

  • brave_video_search

    • Search the web for videos

    • Inputs:

      • query: (string): The term to search for videos

      • count: (number, optional): The number of videos to return (max 20, default 10)

      • offset (number, optional, default 0): The zero-based offset for pagination (max 9)

      • freshness (enum, optional): Filters search results by when they were discovered

        • The following values are supported

          • pd: Discovered within the last 24 hours.

          • pw: Discovered within the last 7 Days.

          • pm: Discovered within the last 31 Days.

          • py: Discovered within the last 365 Days

          • YYYY-MM-DDtoYYYY-MM-DD: Custom date range (e.g., 2022-04-01to2022-07-30)

  • brave_llm_context_search

    • Pre-extracted web content optimized for AI agents, LLM grounding, and RAG pipelines.

    • Uses Brave's balanced context threshold mode in compact mode and disables Brave relevance filtering in full mode.

    • Inputs:

      • query (string): The search query. Maximum 400 characters and 50 words.

      • url (string, optional): Optional URL to target. When provided, query and URL are combined for retrieval and only snippets from this exact URL are returned.

      • count (number, optional, default 8): The maximum number of search results considered. Minimum 1, maximum 50.

      • maximumNumberOfUrls (number, optional, default 8): The maximum number of URLs to include in the response. Minimum 1, maximum 50.

      • maximumNumberOfTokens (number, optional, default 2048): The approximate maximum number of tokens in the returned context. Minimum 1024, maximum 32768.

      • maximumNumberOfSnippets (number, optional, default 16): The maximum number of snippets across all URLs. Minimum 1, maximum 100.

      • maximumNumberOfTokensPerUrl (number, optional, default 512): The maximum number of tokens per URL. Minimum 512, maximum 8192.

      • maximumNumberOfSnippetsPerUrl (number, optional, default 2): The maximum number of snippets per URL. Minimum 1, maximum 100.

      • responseMode (enum, optional, default compact): compact applies Brave's balanced relevance filtering plus local snippet filtering/truncation. full disables Brave's relevance filtering and returns raw snippets without local filtering or truncation.

      • maxSnippetChars (number, optional, default 400): Maximum characters per snippet in compact mode. Minimum 80, maximum 4000.

      • maxOutputChars (number, optional, default 8000): Approximate maximum serialized response size in compact mode. Minimum 1000, maximum 100000.

OpenAI Apps & MCP Apps Support

There is now support for OpenAI Apps and MCP Apps in this MCP Server. When UI mode is enabled for each tool there is a corresponding UI widget that let's you control what gets added to the model's context. See the directions in usage with ChatGPT section.

Configuration

Getting an API Key

  1. Sign up for a Brave Search API account

  2. Choose a plan

  3. Generate your API key from the developer dashboard

Runtime modes

By default the MCP server runs in stdio mode.

BRAVE_API_KEY="your_key_here" npx -y brave-search-mcp

To enable Streamable HTTP mode:

BRAVE_API_KEY="your_key_here" npx -y brave-search-mcp --http

By default the server listens on port 3001. The URL is:

http://0.0.0.0:3001/mcp

There are two configuration modes:

  • Env mode: BRAVE_MCP_CONFIG is unset. Feature toggles come from environment variables exactly as in previous releases.

  • File mode: BRAVE_MCP_CONFIG=/path/to/config.toml is set. The TOML file becomes the single source of truth for feature settings, and overlapping feature env vars are ignored with warnings.

Environment-only settings

These settings are always read from the process environment, regardless of mode:

  • BRAVE_API_KEY (required): Brave Search API key.

  • PORT (optional): HTTP port (default: 3001).

  • HOST (optional): Interface to bind to (default: 0.0.0.0).

  • BRAVE_MCP_CONFIG (optional): Absolute or relative path to a TOML config file for feature settings.

Env mode feature settings

When BRAVE_MCP_CONFIG is not set, these feature env vars are supported:

  • ALLOWED_HOSTS (HTTP mode only): Comma-separated list of allowed hostnames for Host header validation.

    • Example: ALLOWED_HOSTS=localhost,127.0.0.1,my-app.ngrok-free.app

    • Use hostnames only (no scheme/path), e.g. my-app.ngrok-free.app not https://my-app.ngrok-free.app/mcp

  • BRAVE_MCP_POLICY_FILE: JSON policy file path.

  • BRAVE_MCP_POLICY_REDACT: true to redact matched text instead of blocking it.

  • BRAVE_MCP_REQUEST_LIMIT: Positive integer request cap.

  • BRAVE_MCP_WINDOW_SECONDS: Non-negative integer rolling window size.

  • BRAVE_MCP_COOLDOWN_SECONDS: Non-negative integer cooldown after the limit is exceeded.

  • BRAVE_MCP_AUDIT_LOG: true to emit audit logs.

  • BRAVE_MCP_AUDIT_LOG_RAW: true to include raw query text in audit logs.

  • BRAVE_MCP_REQUIRE_JUSTIFICATION: true to reject tool calls without a justification string.

Examples:

# Local only
HOST=127.0.0.1 ALLOWED_HOSTS=localhost,127.0.0.1 BRAVE_API_KEY="your_key_here" npx -y brave-search-mcp --http
# Local with ngrok tunnel
HOST=127.0.0.1 ALLOWED_HOSTS=localhost,127.0.0.1,my-app.ngrok-free.app BRAVE_API_KEY="your_key_here" npx -y brave-search-mcp --http --ui

File mode (BRAVE_MCP_CONFIG)

When BRAVE_MCP_CONFIG is set, the file controls feature configuration, including the HTTP host allowlist.

[auth]
httpApiKey = "sk-..."
requireAuth = true
callerId = "team-a"

[auth.jwt]
jwksUri = "https://idp.example.com/.well-known/jwks.json"
audience = "brave-search-mcp"
clockSkewSeconds = 30

[auth.oauth]
issuer = "https://idp.example.com"
audience = "brave-search-mcp"
clientId = "client-123"
clientSecret = "super-secret"
verifyStrategy = "jwks"

[audit]
enabled = true
logRaw = false
hmacSecret = "audit-secret"

[policy]
file = "/etc/brave-mcp/policy.json"
redact = false

[guardrail]
requestLimit = 100
windowSeconds = 60
cooldownSeconds = 10
requireJustification = false

[server]
allowedHosts = [
  "localhost",
  "127.0.0.1",
  "my-app.ngrok-free.app",
]

Notes:

  • BRAVE_API_KEY, PORT, and HOST remain environment-only even in file mode.

  • If you set overlapping feature env vars such as BRAVE_MCP_REQUEST_LIMIT or ALLOWED_HOSTS alongside BRAVE_MCP_CONFIG, startup warns that they are being ignored.

  • Unknown TOML keys also emit warnings so typos like [guardrails] are visible before you debug runtime behavior.

To validate a config file without starting the server, use the packaged entrypoint after building the app workspace:

pnpm -C apps/brave-search-mcp run build
node apps/brave-search-mcp/dist/index.js --check-config ./apps/brave-search-mcp/test/fixtures/config.valid.toml

To run the built local entrypoint in file mode:

pnpm -C apps/brave-search-mcp run build
BRAVE_API_KEY="your_key_here" BRAVE_MCP_CONFIG="$PWD/apps/brave-search-mcp/test/fixtures/config.valid.toml" node apps/brave-search-mcp/dist/index.js --http

Usage with ChatGPT

The Brave Search MCP Server can be used with the web UI of ChatGPT. It takes a few steps.

1. Enable Developer Mode in ChatGPT

Settings → Apps → Advanced settings → Developer mode

Additional instructions here

2. Run the Brave Search MCP in HTTP mode and UI mode

BRAVE_API_KEY="your_key_here" npx -y brave-search-mcp --http --ui

3. Create a local tunnel to expose the MCP Server to ChatGPT

Sign up and configure ngrok, the free plan works.

ngrok http 3001

Take note of the forwarding URL.

...
Forwarding                    https://john-joe-asdf.ngrok-free.dev -> http://localhost:3001
...

4. Add Brave Search MCP as a Connector to ChatGPT

Open ChatGPT Apps settings

Click Apps

Click Create Apps

Fill out the form using the URL from step 3 as the MCP Server URL, but add /mcp.

https://john-joe-asdf.ngrok-free.dev/mcp

For Authentication, select 'No Auth'

Tick the checkbox for 'I understand and want to continue'

Then click Create.

5. Using the Brave Search MCP Server

In the ChatGPT UI, click the '+' button, scroll to '...more', select the newly created Brave Search app, and enter your query.

Usage with Claude Code

For Claude Code users, run this command:

Windows:

claude mcp add-json brave-search '{"command":"cmd","args":["/c","npx","-y","brave-search-mcp"],"env":{"BRAVE_API_KEY":"YOUR_API_KEY_HERE"}}'

Linux/macOS:

claude mcp add-json brave-search '{"command":"npx","args":["-y","brave-search-mcp"],"env":{"BRAVE_API_KEY":"YOUR_API_KEY_HERE"}}'

Replace YOUR_API_KEY_HERE with your actual Brave Search API key.

Usage with Claude Desktop

MCP Bundle (MCPB)

  1. Download the mcpb file from the Releases

  2. Open it with Claude Desktop or Go to File -> Settings -> Extensions and drag the .mcpb file to the window to install it

Docker

  1. Clone the repo

  2. Build the image from the repo root

docker build -t brave-search-mcp:latest -f apps/brave-search-mcp/Dockerfile .

  1. Run it directly if you want HTTP mode:

docker run --rm -p 3001:3001 -e BRAVE_API_KEY="YOUR_API_KEY_HERE" brave-search-mcp:latest --http

  1. Add this to your claude_desktop_config.json for stdio mode:

{
  "mcp-servers": {
    "brave-search": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "BRAVE_API_KEY",
        "brave-search-mcp"
      ],
      "env": {
        "BRAVE_API_KEY": "YOUR API KEY HERE"
      }
    }
  }
}

NPX

Add this to your claude_desktop_config.json:

{
  "mcp-servers": {
    "brave-search": {
      "command": "npx",
      "args": [
        "-y",
        "brave-search-mcp"
      ],
      "env": {
        "BRAVE_API_KEY": "YOUR API KEY HERE"
      }
    }
  }
}

Usage with LibreChat

Add this to librechat.yaml

brave-search:
  command: sh
  args:
    - -c
    - BRAVE_API_KEY=API KEY npx -y brave-search-mcp

Contributing

Contributions are welcome! See CONTRIBUTING.md for development setup, monorepo structure, and release instructions. For UI build workflow details, including the entrypoint orchestrator, see UI Build Orchestrator.

Disclaimer

This library is not officially associated with Brave Software. It is a third-party implementation of the Brave Search API with a MCP Server.

License

This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.

Available Tools

5 tools

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

  1. 5 tool updatesv1.0.0
    • First observedbrave_image_search
    • First observedbrave_local_search
    • First observedbrave_news_search
    • First observedbrave_video_search
    • First observedbrave_web_search
Deploy Server
One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

TDQS

A4/5.0

Scored across 5 tools

Disambiguation5/5

Each tool has a clearly distinct purpose targeting different content types: images, local businesses, news articles, videos, and general web content. The descriptions explicitly differentiate use cases (e.g., 'near me' for local search, 'recent events' for news), eliminating any ambiguity between tools.

Naming Consistency5/5

All tool names follow a consistent 'brave_[content_type]_search' pattern, using snake_case uniformly. This predictable naming scheme makes it easy to identify the tool's function and ensures no confusion from mixed conventions.

Tool Count5/5

With 5 tools, this server is well-scoped for its purpose of providing specialized search capabilities across different content types. Each tool earns its place by covering distinct search domains, avoiding both redundancy and excessive fragmentation.

Completeness5/5

The tool set comprehensively covers the search domain by addressing all major content types: images, local information, news, videos, and general web content. There are no obvious gaps, as agents can handle any search query by selecting the appropriate specialized or general tool.

Maintenance

ActivityInactive
ResponsivenessUnresponsive

Resources

Related MCP Connectors

Related MCP Servers

Appeared in Searches