Jan AI MCP web search guide
How to Add Web Search to Jan AI with Webstractor MCP
Jan can give local and remote models tools through a visual MCP settings screen. Its current releases also include native search and fetch, so this integration is not about claiming Jan lacks web access. It is about giving teams a consistent hosted search tool, schema, cache behavior, and source-link workflow across multiple agent clients.
Quick answer
Add Webstractor as an HTTP MCP server in Jan.
Open Settings → MCP Servers, click Add MCP Server, choose HTTP, enter https://webstractor.com/mcp, and enable the server. Jan now includes native search and fetch tools too; use Webstractor when you want the same source-linked, provider-neutral search contract across Jan and your other MCP clients.
Server name: Webstractor
Transport: HTTP
URL: https://webstractor.com/mcpBefore you begin
What you need
- A current Jan Desktop installation
- A model that supports tool calls
- Outbound HTTPS access
Step-by-step setup
Connect Webstractor to Jan AI
Open Jan’s MCP settings
Go to Settings → MCP Servers. Jan manages remote servers in its interface, so you do not need to locate or hand-edit an application config file.
Add a remote HTTP server
Click + Add MCP Server. Name it Webstractor, choose HTTP—the Streamable HTTP option, not SSE—and enter the endpoint below.
name: Webstractor
transport: HTTP
url: https://webstractor.com/mcpEnable and inspect the connection
Toggle the server on. Jan should show a green status indicator. Keep permission prompts enabled while you validate the tool behavior.
Use a model with tool calling enabled
Select a compatible model and ensure its provider or model settings permit tools. A connected server cannot compensate for a model that does not produce tool calls.
Test explicit search routing
Prompt: “Use Webstractor search_web to find five current sources about MCP Streamable HTTP. Return a one-sentence relevance note for each URL.” This distinguishes Webstractor from Jan’s native search during setup.
Tool overlap
When to use Webstractor instead of Jan’s native search
Jan 0.8.4 introduced native search and fetch, which may be the simplest choice for a Jan-only workflow. Webstractor is useful when your prompt and output need to behave consistently in Jan, IDE agents, and automation clients.
Naming the desired server and tool in important prompts prevents ambiguous routing when both search capabilities are enabled.
Routing
Keep the active tool set small
Jan offers Smart MCP tool routing when several servers are connected. Enable it if tool definitions consume too much context, but first verify Webstractor with a direct call so routing changes do not hide a setup problem.
- Start with Webstractor as the only search MCP
- Validate search_web explicitly
- Enable Smart MCP routing as the server list grows
- Use a lightweight routing model if appropriate
Troubleshooting
Fix green-status and tool-selection problems separately
No green indicator usually points to the URL, transport, firewall, or a temporary connection problem. A green server whose tools are never called usually points to the selected model, permissions, or competing web tools.
Jan’s default MCP tool timeout is 30 seconds. Keep the first query short before increasing it.
Available data
What you can extract
- Up to 10 ranked results per web query
- Titles, snippets, and original source links
- Site-restricted discovery for documentation research
- Markdown and structured JSON result modes
AI workflows
Where normalized data helps
- Reuse one search tool across Jan and coding agents
- Research current libraries with traceable sources
- Locate PDF sources before extracting them
- Build provider-neutral agent prompts
Boundaries
Public data only
- Jan’s native web tools and Webstractor can overlap; enable only the tools your workflow needs.
- Tool permission settings control approval, not result accuracy.
- Search does not replace source review for legal, medical, or financial decisions.
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
Jan AI and Webstractor FAQ
Does Jan already have web search?
Yes. Current Jan releases include native web search and fetch. Webstractor adds a portable MCP tool and a consistent source-linked output contract across clients.
Should I choose HTTP or SSE?
Choose HTTP in Jan. That label represents the modern Streamable HTTP transport used by the Webstractor endpoint.
Why is the server green but unused?
Confirm the model supports tools, keep approval prompts visible, and explicitly request Webstractor search_web in the test prompt.
Ready to try it?