Single fetch tool discovered
The server exposed exactly one tool, fetch, with url, max_length, start_index, and raw parameters.
evidence/trust5/2026-06-13-mcp-pilot.json
MCP serverPublished
MCP Fetch Reference Server
The Fetch reference server passed a real MCP smoke test: it fetched a live URL and returned simplified markdown with length controls. The score is good for a focused, single-purpose tool, with caution because it will fetch arbitrary URLs and needs client-side egress and internal-network controls.
Independent trust badge
The visible trust mark for this verdict.
Badge clicks resolve to this canonical verdict so the score, test date, evidence, limitations, and reply status remain attached.
Embed
Show this badge on your site
[](https://silentcritique.com/verdicts/mcp-fetch-reference)Markdown works in GitHub READMEs. The badge always links back to this verdict.
Editorial notice
This page reflects SilentCritique's independent editorial opinion based on the specific test evidence shown. It is not an allegation of unlawful, malicious, fraudulent, or bad-faith conduct. SilentCritique does not accept payment to remove criticism, change a score, suppress a verdict, or improve an outcome.
Claim tested
Can the public Fetch MCP server retrieve a web page over MCP and return usable, length-bounded text?
Evaluator panel
Protocol harnessSafety reviewerOperator skeptic
Evidence reviewed
Live page fetched and simplified
Fetching https://example.com/ returned its text converted to markdown rather than raw HTML.
evidence/trust5/2026-06-13-mcp-pilot.json
Length bounding worked
The max_length parameter capped the returned content, with start_index available for paginated continuation.
evidence/trust5/2026-06-13-mcp-pilot.json
Test setup
- Started mcp-server-fetch via uvx over MCP stdio using the official client SDK.
- Listed tools and called fetch on https://example.com/ with max_length set to 500.
- Stored the full tool-call evidence in evidence/trust5/2026-06-13-mcp-pilot.json.
Strengths
- The tool surface is minimal and easy for a client to reason about.
- HTML-to-markdown simplification keeps responses compact and model-friendly.
- max_length and start_index give explicit control over response size.
Failure modes
- The tool will fetch arbitrary URLs, so without client controls it can reach internal network addresses (SSRF).
- A raw mode returns unsimplified HTML, which can flood a model context if used carelessly.
- Autonomous fetch behavior depends on robots.txt handling that a client can override.
What would improve the score
- Document a recommended URL allowlist or denylist pattern for clients.
- Provide a built-in option to block private and link-local address ranges.
- Surface robots.txt and redirect behavior more prominently in tool descriptions.
Limitations
- This was an unsolicited smoke test of the public package, not a full security audit.
- Only a single public URL on macOS was tested.
- The test did not probe internal-address or redirect handling.
Visible dissent
- The protocol harness scored this well because the single tool did exactly one thing correctly.
- The safety reviewer withheld points because unrestricted URL fetching is an egress risk that the server delegates entirely to the client.
Right of reply
No vendor reply has been requested or published as of 2026-06-13. SilentCritique will publish factual corrections or a right of reply through the corrections process.
Methodology matters
Scores are only meaningful when the rubric, date, evidence, and dissent are visible.
Read methodology