fetch: hardcoded 30s timeout with no CLI/env/per-request override

[chirag127]

Bug: fetch MCP server has hardcoded 30s timeout with no CLI/env/per-request override

Problem

The fetch MCP server has a hardcoded 30-second request timeout. This is:

1. Too short for large file downloads (GitHub raw content > 10MB, PDF documentation sites, slow API endpoints)
2. Too long for health checks and fast lookups where a 5s timeout would surface problems faster
3. Not overridable — there is no CLI flag, environment variable, or per-request option to change the timeout

Agents that use fetch for slow operations (downloading a large WASM binary for inspection, fetching a paginated API with slow backend) hit the 30s wall and get a timeout error with no way to extend it for that specific call.

Expected behavior

Make the timeout configurable at three levels (most-specific wins):

1. Per-request: include timeout_ms in the tool call arguments

   { "url": "https://example.com/large-file.pdf", "timeout_ms": 120000 }

2. Environment variable for server-wide default: FETCH_TIMEOUT_MS=60000
3. CLI flag: --timeout 60000 when starting the server

The current 30s hardcoded value becomes the default if none of the above are set.

Why

Different use cases need different timeouts. Documentation lookups: 5-10s. File downloads: 60-300s. Streaming responses: potentially unlimited with streaming support. Hardcoding one value for all cases means either slow failures on legitimate long operations or slow detection of actually-down endpoints.

Environment

• MCP servers: @modelcontextprotocol/server-fetch (latest)
• Runtime: Node.js v24.18.0, Windows 11 Enterprise
• Hit when: fetching large GitHub raw files and slow public API endpoints via fetch MCP tool