- Two tools
web_search: find relevant URLs for a queryweb_fetch: fetch latest page content (markdown/text)
- Cost-aware workflow: recall → search → selective fetch (save quota)
- Docs-first answers: answers can cite freshly fetched official pages
Official reference: Ollama Web Search
Note
Prerequisites: Node.js for npm install. OLLAMA_API_KEY is required for web_search.
npm:
npm install -g @neabyte/websearch-mcpRun the server (example for MCP clients using stdio):
websearch-mcpIf you want to set the API key for Ollama web search:
export OLLAMA_API_KEY="..."
websearch-mcpYou can integrate this MCP server directly into editors (Cursor/VSCode) via stdio MCP configuration. Two common options:
- Global install:
npm install -g @neabyte/websearch-mcpthen runwebsearch-mcp - No global install: use
npx -y @neabyte/websearch-mcp(recommended for editors)
Paste this MCP server config in your editor (example JSON used by many MCP clients):
{
"mcpServers": {
"websearch-mcp": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@neabyte/websearch-mcp"],
"env": {
"OLLAMA_API_KEY": "YOUR_OLLAMA_API_KEY"
}
}
}
}
OLLAMA_API_KEYis required forweb_search.web_fetchcan run without a key.
After the server is registered in your editor, try a prompt like this:
Use installed MCP Servers: `websearch-mcp` - `web_fetch` to fetch https://reactnative.dev/docs/environment-setup. Summarize the macOS install steps.
Direct fetch (when the URL is known):
Use installed MCP Servers: `websearch-mcp` - `web_fetch` to fetch the official React Native documentation pages. Question: How to install in macOS?
Search → fetch (when the URL is unknown):
Use installed MCP Servers: `websearch-mcp` - `web_search` to find official React Native docs URLs for "environment setup macOS". Then use `web_fetch` on the most relevant official doc page URLs. Question: How to install React Native in macOS?
This approach uses two steps:
- Recall: the LLM proposes likely official URLs from its knowledge
- Fetch: the LLM calls
web_fetchto retrieve the latest content
This is useful for fast-changing documentation, because the final answer is grounded in pages fetched at tool-call time.
Ollama can get expensive if you web_fetch many pages. A cost-aware pattern is:
- Recall first: ask for a shortlist of likely official URLs
- Search if needed: use
web_searchto find fresh candidates - Fetch selectively: fetch only 1–3 most relevant pages
With this pattern, you only spend fetch calls on pages you actually cite.
| Prompt | Output |
|---|---|
![]() |
![]() |
- Prompt: ask the model to use
websearch-mcp+web_fetchto fetch official React Native docs - Output: the model fetches
reactnative.devand summarizes install steps
Even if web_search has its own quota/limits, the “double tool” pattern can reduce total cost by
shortlisting URLs first and fetching only what you need.
Use websearch-mcp web_search with query: "latest crypto payment gateway 2026".
Then use websearch-mcp web_fetch to fetch the official pages from the top results.
Answer with a short list of the latest crypto payment gateways and include the source links.
From the repo root (requires Deno).
Check — format, lint, and typecheck source:
deno task checkUnit Tests:
deno task test- Tests live under
tests/
This project is licensed under the MIT license. See the LICENSE file for details.


