Kilo Code MCP web search / 2026
How to Add Web Search to Kilo Code with MCP in 2026
Kilo Code does not ship with MCP servers preinstalled, but its extension and CLI can connect to remote servers. Webstractor adds a focused discovery tool without a local process or a search-provider key. Kilo’s permission system also lets you keep the tool confirm-first or approve only the read-only search action.
Quick answer
Add Webstractor as a Remote (HTTP) MCP server.
In Kilo Code, open Settings → MCP, add a Remote (HTTP) server, and enter the hosted endpoint. Kilo tries Streamable HTTP first. Verify the server status, then approve search_web when the agent requests it.
{
"mcp": {
"webstractor": {
"type": "remote",
"url": "https://webstractor.com/mcp",
"enabled": true,
"timeout": 15000
}
}
}Before you begin
What you need
- Current Kilo Code extension or CLI
- A tool-calling model
- Access to Settings → MCP or a Kilo config file
Step-by-step setup
Connect Webstractor to Kilo Code
Add a Remote (HTTP) server
In the VS Code extension, open Settings → MCP, click Add Server, and choose Remote (HTTP). Name it webstractor, enter the endpoint, and leave headers empty. Kilo attempts Streamable HTTP before its legacy SSE fallback.
Use project or global JSON when preferred
Put project configuration in kilo.jsonc or .kilo/kilo.jsonc. Use ~/.config/kilo/kilo.jsonc only when every workspace should have the tool.
{
"mcp": {
"webstractor": {
"type": "remote",
"url": "https://webstractor.com/mcp",
"enabled": true,
"timeout": 15000
}
}
}Approve and verify one search
Ask for explicit tool use. When Kilo’s Permission Dock appears, approve once for testing. Use Approve Always only if you want this read-only tool to run without future confirmation.
# Paste into Kilo Code
Use webstractor search_web to find the official Kilo Code MCP configuration docs. Return 5 source URLs, then identify the most authoritative result.Permissions
Keep approval narrow
Kilo names MCP permission keys as `{server}_{tool}`. If you want automatic approval for only web search, allow the specific `webstractor_search_web` key instead of a wildcard that covers every Webstractor tool.
The default confirmation is useful while evaluating model behavior. It shows the arguments—query, result limit, and filters—before the request runs.
{
"permission": {
"webstractor_search_web": "allow"
}
}Troubleshooting
Read Kilo’s server status first
A failed server indicator usually points to the endpoint, timeout, or network. Confirm that the entry is enabled and uses type `remote`. If a valid server is connected but the tool never runs, check whether its permission is denied.
Project configuration takes precedence over global configuration. Remove or fix a stale project entry if it shadows the working global server.
Available data
What you can extract
- Up to 10 relevant public pages
- Titles, canonical URLs, and result snippets
- Language, country, site, and PDF filters
- Markdown or typed JSON result feeds
AI workflows
Where normalized data helps
- Research an unfamiliar error before patching
- Find official docs for a dependency
- Locate implementation examples on a specific domain
- Discover public PDF specifications
Boundaries
Public data only
- Kilo asks for tool approval unless permissions say otherwise.
- Search snippets should not be treated as the full source.
- Public search does not include private repositories or authenticated documents.
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
Kilo Code and Webstractor FAQ
Does Kilo Code support Webstractor directly?
Yes. Kilo supports remote HTTP MCP servers and tries Streamable HTTP first.
Do I need to auto-approve web search?
No. The default confirm flow works; auto-approval is an optional convenience for a trusted read-only tool.
Where should I store the config?
Use .kilo/kilo.jsonc for one repository or ~/.config/kilo/kilo.jsonc for all projects.
Ready to try it?