Best fetch MCP servers
We tested 10 fetch servers; 8 are ranked below. 2 aren't ranked: no passing tool call and no GitHub star count found; no passing tool call and under 50 GitHub stars. Short answer: MarkItDown (Microsoft) (100/100). This list includes paid and keyed servers; servers that need a key are labelled, and we have not tested any of them with real credentials.
Last tested 2026-10-27
How we test: methodology. Context cost: lowest context cost MCP servers. Not tested by us: security, output quality, and behaviour with real credentials.
Ranked by test score
| Server | Score | Status | Tools | Schema tokens (est.) | Functional tasks | GitHub stars |
|---|---|---|---|---|---|---|
MarkItDown (Microsoft)markitdown-mcp | 100 | Working | 1 | ~63 | fetch public URL (https://example.com/): pass (convert_to_markdown) | 191,528 |
Fetcher MCPfetcher-mcp | 80 | Working | 3 | ~709 | fetch public URL (https://example.com/): pass (fetch_url) | 1,088 |
Fetch (kazuph)@kazuph/mcp-fetch | 75 | Working | 1 | ~1,496 | fetch public URL (https://example.com/): pass (imageFetch) | 41 |
npx fetch@tokenizin/mcp-npx-fetch | 75 | Working | 4 | ~258 | fetch public URL (https://example.com/): pass (fetch_html) | 42 |
Apify (official)@apify/actors-mcp-server needs key: not tested with credentials | 70 | Needs key | Not measured | Not measured | Not measured | 12,499 |
Fetch (reference, Python)mcp-server-fetch | 70 | Blocked by our test setup | 1 | ~224 | fetch public URL (https://example.com/): not run (fetch) | 91,469 |
Read Website Fast@just-every/mcp-read-website-fast | 70 | Blocked by our test setup | 1 | ~127 | fetch public URL (https://example.com/): not run (read_website) | 163 |
YouTube Transcript@kimtaeyoon83/mcp-server-youtube-transcript | 70 | Connects · tools listed | 1 | ~79 | fetch public URL: not run (-) | 604 |
Tested, not ranked
- YouTube Transcript (Python) (
mcp-youtube-transcript): no passing tool call and no GitHub star count found. - Fetch (Python alt) (
mcp-server-fetch-python): no passing tool call and under 50 GitHub stars.
Listings with no passing tool call and under 50 GitHub stars aren't ranked here unless a check against the vendor's own docs confirmed them. They stay in the full directory.
Ranked by score, then most recent activity, then more tools, then name. Failed = did not complete the MCP handshake without credentials or a backing service.
Small-model runs for MarkItDown (Microsoft) (directional, 3 runs each): qwen2.5:3b-instruct-q4_K_M completed 0% of tasks; qwen2.5:1.5b-instruct-q4_K_M completed 100% of tasks. MCP with local models.
FAQ
What is the best fetch MCP server?
In our 2026-10-22 tests, MarkItDown (Microsoft) (markitdown-mcp) ranked first with 100/100 and 1 tool listed. Ranking: score, then most recent activity, then more tools, then name.
Which fetch MCP server uses the least context?
MarkItDown (Microsoft) lists 1 tool at about 63 schema tokens, the lowest among maintained servers here that we measured.
Which fetch MCP servers are deprecated?
None of the 10 we tested, as of 2026-10-27.
Why are only 8 of the 10 tested fetch servers ranked?
We tested 10 fetch servers; 8 are ranked below. 2 aren't ranked: no passing tool call and no GitHub star count found; no passing tool call and under 50 GitHub stars. Not ranked: YouTube Transcript (Python) (no passing tool call and no GitHub star count found); Fetch (Python alt) (no passing tool call and under 50 GitHub stars).