---
title: "Brave Search MCP: Web Research"
source: https://docs.autohand.ai/guides/mcp/brave-search
---

# Brave Search MCP: Web Research

Give Autohand the ability to search the web. Brave Search MCP lets the agent look up error messages, find documentation, and research solutions in real time without you switching to a browser.

## What Brave Search does

[Brave Search MCP](https://github.com/brave/brave-search-mcp-server) connects Autohand to the Brave Search API. The agent can perform web searches, read summaries of results, and use that information to solve problems. This is useful when the agent encounters an unfamiliar error, needs to check if a package exists, or wants to find community solutions for a specific issue.

Unlike documentation-specific tools, Brave Search covers the entire web: Stack Overflow answers, GitHub issues, blog posts, release notes, and more.

## Setup

Brave Search MCP uses the stdio transport. You need a Brave Search API key (free tier available at [brave.com/search/api](https://brave.com/search/api/)).

``` bash
# Add Brave Search MCP
autohand mcp add brave-search npx -y @anthropic/mcp-server-brave-search

# Set your API key as an environment variable
export BRAVE_API_KEY="your-brave-api-key"

# Verify the connection
autohand mcp list
```

Or configure it in `~/.autohand/config.json`:

``` json
{
  "mcp": {
    "servers": [
      {
        "name": "brave-search",
        "transport": "stdio",
        "command": "npx",
        "args": ["-y", "@anthropic/mcp-server-brave-search"],
        "env": {
          "BRAVE_API_KEY": "your-brave-api-key"
        },
        "autoConnect": true
      }
    ]
  }
}
```

## Usage examples

The agent gets access to `mcp__brave-search__brave_web_search` and uses it when it needs external information.

### Debug an unfamiliar error

``` bash
# Paste an error and let the agent research it
autohand --prompt "I'm getting 'ERR_OSSL_EVP_UNSUPPORTED' when running my Node.js app. What's causing this and how do I fix it?"

# The agent searches for the error, finds that it's related to OpenSSL 3.0,
# and provides the fix
```

### Find the right package

``` bash
# Discover packages for a specific task
autohand --prompt "Find the best Node.js library for generating PDF invoices with custom templates"

# Brave Search finds and compares options like puppeteer, pdfkit, jsPDF
```

### Check for breaking changes

``` bash
# Research before upgrading
autohand --prompt "What breaking changes are in TypeScript 5.7? I'm upgrading from 5.5."

# Searches for release notes and migration guides
```

### Research a pattern or approach

``` bash
# Get current best practices
autohand --prompt "What's the recommended way to handle authentication in Next.js App Router in 2026?"

# Finds recent blog posts and official recommendations
```

## Combining with other MCPs

Brave Search works well alongside other MCP servers:

-   **Brave Search + [Context7](/docs/guides/mcp/context7-docs.html)**: Use Brave to find the right library, then Context7 to get its full API docs.
-   **Brave Search + [Firecrawl](/docs/guides/mcp/firecrawl-scraping.html)**: Use Brave to locate a relevant page, then Firecrawl to extract its full content as clean markdown.

``` json
{
  "mcp": {
    "servers": [
      {
        "name": "brave-search",
        "transport": "stdio",
        "command": "npx",
        "args": ["-y", "@anthropic/mcp-server-brave-search"],
        "env": { "BRAVE_API_KEY": "your-key" },
        "autoConnect": true
      },
      {
        "name": "context7",
        "transport": "http",
        "url": "https://mcp.context7.com/mcp",
        "headers": { "CONTEXT7_API_KEY": "your-key" },
        "autoConnect": true
      }
    ]
  }
}
```

## Tips

-   The free Brave Search API tier gives you 2,000 queries per month, which is enough for most development workflows.
-   Be specific in your prompts. "Fix my React error" will produce worse search results than "Fix React hydration mismatch error in Next.js 15".
-   The agent decides when to search on its own. You do not need to tell it to "search the web" explicitly.
-   Search results include snippets and URLs. The agent can use [Firecrawl MCP](/docs/guides/mcp/firecrawl-scraping.html) to read the full page content when snippets are not enough.