Roo Code MCP web search guide

How to Add Web Search to Roo Code with Webstractor MCP

Roo Code does not bundle MCP servers, but it supports modern remote Streamable HTTP connections at global and project scope. A project-scoped Webstractor entry is useful when current documentation research is part of the repository workflow and teammates need a visible, repeatable tool definition.

20264 min readFor agent builders

Quick answer

Add Webstractor to .roo/mcp.json with streamable-http.

Open Roo Code’s MCP settings, choose Edit Project MCP, and add Webstractor with type streamable-http and the hosted /mcp URL. Project configuration makes the tool discoverable for that repository, while the global mcp_settings.json is better for a personal default across workspaces.

MCP configuration

{
  "mcpServers": {
    "webstractor": {
      "type": "streamable-http",
      "url": "https://webstractor.com/mcp",
      "disabled": false
    }
  }
}

Before you begin

What you need

Roo Code installed in VS Code
MCP Servers enabled
A tool-capable model and active workspace

Step-by-step setup

Connect Webstractor to Roo Code

Choose project or global scope

Use .roo/mcp.json when the server belongs to one repository and mcp_settings.json when you want it in every workspace. A project server with the same name takes precedence.

Open the matching config from Roo

Click the MCP icon in Roo Code, scroll down, and select Edit Project MCP or Edit Global MCP.

Declare a Streamable HTTP server

Roo Code requires the hyphenated transport value streamable-http. Add this under mcpServers and keep it enabled.

.roo/mcp.json

{
  "mcpServers": {
    "webstractor": {
      "type": "streamable-http",
      "url": "https://webstractor.com/mcp",
      "disabled": false
    }
  }
}

Inspect permissions and tool list

Confirm search_web is listed. Do not mark it Always allow until you have reviewed its inputs and your team’s MCP policy.

Test in a research-oriented mode

Prompt: “Use search_web with site=developers.cloudflare.com to find current Cache API documentation. Do not edit files; give me the best five sources and a source-backed implementation outline.”

Configuration scope

Project MCP makes research dependencies visible

A checked-in .roo/mcp.json tells contributors that the repository expects a particular external tool. Because the Webstractor endpoint works anonymously, the shared entry does not need a credential.

If account usage is later added through a header, keep that personal configuration out of the repository and use an appropriate secret mechanism.

Mode-aware use

Research in Architect mode before changing code

Roo’s modes make it natural to separate evidence gathering from implementation. Use search_web while reasoning about options, inspect source URLs, then switch to a code-changing mode after the plan is accepted.

Search with a narrow technical query
Prefer an official site filter
Open or extract the winning source
Apply code changes only after review

Troubleshooting

Do not omit Roo’s transport type

For URL-based configurations, Roo Code cannot infer the transport. Set type to streamable-http; a bare url produces an error.

If the server is connected but absent from prompts, verify Enable MCP Servers is checked and that search_web is not listed in disabledTools.

Available data

What you can extract

Current public-web result lists
Original URLs for evidence review
Optional official-domain filtering
Compact context suitable for planning before implementation

AI workflows

Where normalized data helps

Research framework behavior in Architect mode
Find official references before debugging
Compare recent public solutions without opening a browser session
Discover source pages for later extraction

Boundaries

Public data only

A committed .roo/mcp.json must never contain a live secret.
Search results can be stale or wrong and should be checked at the source.
Auto-approval requires both global and per-tool permission; leave it off initially.

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

Roo Code and Webstractor FAQ

Should I use .roo/mcp.json or global settings?

Use project config for a repository-specific shared tool definition and global settings for your personal cross-workspace default.

Can Roo Code auto-approve search_web?

Yes, but global MCP auto-approval and the per-tool Always allow option must both permit it. Manual approval is the safer starting point.

Why does a URL-only config fail?

Roo requires an explicit type for remote servers. Use "streamable-http" for Webstractor.

Ready to try it?

Turn a public URL into useful context.

Configure Roo Code search Open Webstractor MCP docs