BoltAI MCP web search / 2026
How to Add Web Search to BoltAI with MCP in 2026
BoltAI can attach MCP tools to individual agents, which is ideal for a research profile that needs current sources without enabling web access everywhere. BoltAI 2.x supports remote connectors and per-tool selection; older plugin documentation uses a local stdio configuration. This guide covers the modern route and a compatible fallback.
Quick answer
Use a current custom connector or the documented mcp.json bridge.
Current BoltAI releases support custom remote MCP connectors. Add the Webstractor endpoint in Settings → Connectors, enable its tools for the agent, and test search_web. If your BoltAI build exposes only the Plugins JSON editor, use the documented mcp-remote configuration.
{
"mcpServers": {
"webstractor": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://webstractor.com/mcp"]
}
}
}Before you begin
What you need
- Current BoltAI for macOS
- A tool-capable provider or local model
- Node.js only if using the legacy plugin bridge
Step-by-step setup
Connect Webstractor to BoltAI
Prefer a custom remote connector
Open Settings → Connectors, add a custom MCP server, name it webstractor, and use the hosted endpoint. BoltAI 2.2 introduced custom remote connectors; no authentication is needed for Webstractor’s public allowance.
Use mcp.json on plugin-only versions
If your version has Settings → Plugins rather than custom connectors, click the plus button, edit the MCP configuration, save, and reload. BoltAI stores this file at ~/.boltai/mcp.json.
{
"mcpServers": {
"webstractor": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://webstractor.com/mcp"]
}
}
}Enable search for one agent
In the agent’s plugin or tool dropdown, enable webstractor and select search_web. BoltAI 2.x supports individual tool selection, so the research agent does not need every server tool.
# Paste into BoltAI
Use webstractor search_web to find the current BoltAI MCP server documentation. Give me 5 URLs with one-line relevance notes.Agent design
Create a focused research profile
Attach search_web and extract_public_url to a dedicated research agent, then require source links in its instructions. Other BoltAI agents can remain tool-free or use different connectors.
This separation reduces accidental tool calls and makes it clear when an answer is based on current public sources rather than model memory.
- Enable search_web for discovery
- Enable extract_public_url for full-page reading
- Keep finance or media tools off unless the agent needs them
Troubleshooting
Use the path that matches your BoltAI version
Recent BoltAI versions put remote MCP servers under Connectors. The older MCP guide uses Settings → Plugins and a local mcp.json. Update BoltAI if a direct custom connector is unavailable and you do not want the Node.js bridge.
If the server appears but no call runs, confirm that the specific search_web tool is enabled for the active agent—not merely installed globally.
Available data
What you can extract
- Ordered web results with source links
- A configurable result limit up to 10
- Site, language, country, and PDF filters
- Readable Markdown or typed JSON feeds
AI workflows
Where normalized data helps
- Create a research-only BoltAI agent
- Find current product documentation
- Collect sources before drafting in another app
- Search public PDFs and standards
Boundaries
Public data only
- Enable only the tools the agent needs.
- Search output is discovery metadata until the agent reads a result.
- MCP configuration can contain sensitive values; BoltAI does not sync local mcp.json automatically.
webstractor.com does not bypass CAPTCHAs, login walls, paywalls, access controls, or regional restrictions. Review the source’s terms and applicable law before collecting or reusing data.
Common questions
BoltAI and Webstractor FAQ
Does BoltAI support remote MCP servers?
Yes. BoltAI 2.x supports remote connectors, including custom MCP servers; older builds can use mcp-remote through mcp.json.
Can I enable only web search?
Yes. Current BoltAI versions allow selecting individual tools from a server for an agent.
Why is the plugin installed but unavailable in chat?
Enable the server or its search_web tool in the active agent’s plugin dropdown.
Ready to try it?