Continue MCP web search guide
How to Add Web Search to Continue.dev with Webstractor MCP
Continue can connect remote MCP tools to a coding workflow without running another local process. Webstractor gives Agent mode a focused discovery step for current public documentation, issue discussions, and technical references before the model proposes repository changes.
Quick answer
Add a streamable-http block to Continue config.yaml.
Open your Continue configuration and add Webstractor under mcpServers with type streamable-http and URL https://webstractor.com/mcp. After Continue reloads the config, use Agent mode and explicitly ask for search_web during the first test.
mcpServers:
- name: Webstractor
type: streamable-http
url: https://webstractor.com/mcpBefore you begin
What you need
- A current Continue extension
- A model configured for Agent mode and tool use
- Access to Continue config.yaml
Step-by-step setup
Connect Webstractor to Continue.dev
Open the active Continue configuration
Edit the config.yaml used by your current Continue assistant. Preserve existing models, rules, and context providers.
Add the remote MCP block
Continue uses the hyphenated streamable-http transport value. Insert Webstractor under the existing mcpServers list.
mcpServers:
- name: Webstractor
type: streamable-http
url: https://webstractor.com/mcpReload Continue and enter Agent mode
Allow Continue to reload its configuration, switch to Agent mode, and inspect the available tools for search_web.
Run an official-docs search
Prompt: “Use search_web with site=docs.continue.dev and limit=5 to find current MCP configuration documentation. Cite each source and do not edit the workspace.”
Turn the result into a reviewed plan
Select the relevant source, ask Continue to explain how it affects the repository, and approve implementation only after checking the original URL.
Configuration alternative
YAML is clearest for one shared assistant
Continue can also discover JSON MCP files placed under .continue/mcpServers, which helps when migrating configurations from other clients. For a native Continue setup, the YAML block makes the transport and URL explicit.
Keep the server definition small. Tool descriptions arrive from the server, so duplicating operational instructions in config is unnecessary.
Efficient research
Use a domain filter before broad search
Coding questions often have an authoritative documentation domain. Passing site reduces noisy tutorials and keeps a smaller result set in the model context.
- Start with an official hostname
- Ask for three to five results
- Inspect publication or version context
- Extract the selected page if snippets are insufficient
Troubleshooting
Confirm the block is nested under mcpServers
A valid YAML fragment at the wrong indentation can load without the server appearing where expected. Check the active assistant config and exact indentation.
If the tool is present but unused, switch to Agent mode and name search_web. If the server is absent, confirm the URL includes /mcp and type is streamable-http.
Available data
What you can extract
- Ordered search results with original URLs
- Optional site filters for official docs
- Concise snippets for source selection
- Markdown or JSON responses for different agent tasks
AI workflows
Where normalized data helps
- Find current framework documentation before implementing
- Research an error across official and community sources
- Locate upstream changelogs for dependency upgrades
- Discover pages to fetch in a second tool call
Boundaries
Public data only
- MCP tools are available in Agent mode, not every Continue interaction mode.
- Search output must be evaluated before code changes.
- Never place private code or credentials in a public search query.
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
Continue.dev and Webstractor FAQ
Can Continue import JSON MCP config?
Yes. Continue documents automatic discovery of JSON files in .continue/mcpServers. This guide uses native YAML because it is direct and easy to audit.
Why is search_web unavailable in chat?
Use Agent mode with a model that supports tool calls, then verify the MCP server loaded from the active configuration.
Can I restrict results to one site?
Yes. Ask search_web to pass the site hostname, for example docs.continue.dev.
Ready to try it?