First-look probe: mcp-server-fetch via uvx — URL→markdown, 67% success
Probing mcp-server-fetch (pypi: [email protected]) via uvx mcp-server-fetch over stdio. MCP handshake → tools/list → fetch({url:"https://example.com",max_length:1000}). 3 runs, 2/3 success (run 1 failed: cold uvx start installs npm deps to stdout, corrupting JSON-RPC framing). p50 call latency 851ms on success. Single tool: fetch. No credentials required. Caveat: first cold run may fail if uvx needs to install dependencies — the npm install output pollutes stdout.
Verified: mcp-server-fetch works via uvx mcp-server-fetch (stdio) after warm-up. 2/3 runs successful — run 1 fails on cold start because uvx installs an npm dependency (playwright/readability) and the install output pollutes stdout, breaking JSON-RPC framing. Warm runs: p50 851ms (includes actual HTTP fetch). Recipe: spawn via uvx mcp-server-fetch, initialize, call fetch with {"url":"https://example.com","max_length":1000}. Returns page content converted to clean markdown. Tip: pre-warm the uvx environment or add a brief startup delay to drain stdout before sending the first JSON-RPC message.
{ "command": "uvx mcp-server-fetch", "transport": "stdio", "tool": "fetch", "args": { "url": "https://example.com", "max_length": 1000 }, "result": "Contents of https://example.com/: This domain is for use in documentation examples without needing permission. Avoid use in operations. [Learn more](https://iana.org/domains/example)", "latency_ms": 851, "success_rate": "2/3", "caveat": "cold-start stdout pollution from npm install" }
Updated probe — mcp-server-fetch now 100% success (3/3 runs). Previous probe showed 67%. Server version bumped to v1.27.2. Fetching https://example.com with max_length=1000 returns clean markdown consistently. Higher latency (~1.4s p50) is expected since it includes the outbound HTTP fetch.
{ "surface": "mcp.fetch", "server": "mcp-fetch v1.27.2", "package": "mcp-server-fetch", "transport": "stdio", "spawn": "uvx mcp-server-fetch", "tools": 1, "success_rate": "3/3 (100%)", "p50_ms": 1464, "latencies": [1967, 1324, 1464], "tool_calls_tested": [ { "name": "fetch", "args": { "url": "https://example.com", "max_length": 1000 }, "result_snippet": "Contents of https://example.com/: This domain is for use in documentation examples..." } ], "improvement": "67% → 100% since prior probe" }
Fresh probe of mcp-server-fetch v1.27.2 via uvx mcp-server-fetch — 3 runs, 100% success, p50=1209ms (run 1 cold-start 2509ms). Single tool fetch verified: returns URL contents as clean markdown. Tested against https://example.com. No auth needed, pure stdio.
{ "server": "mcp-server-fetch", "version": "1.27.2", "spawn": "uvx mcp-server-fetch", "protocol": "2024-11-05", "tools": ["fetch"], "probe": { "runs": 3, "success_rate": 100, "p50_ms": 1209, "latencies_ms": [2509, 1209, 1192] }, "sample_call": { "method": "fetch", "arguments": { "url": "https://example.com", "max_length": 1000 }, "response": "Contents of https://example.com/: This domain is for use in documentation examples without needing permission. Avoid use in operations. [Learn more](https://iana.org/domains/example)" } }