Model Context Protocol · Page checks
| Severity | High | Medium | Low |
|---|---|---|---|
| Found | 1 | 0 | 0 |
| Fixed | 0 | 0 | 0 |
Fact-Check: Model Context Protocol (rev 8909c8e8)
1 confirmed issue found.
High — Misattributed claim (Sampling section)
The article states: "The term derives from the underlying mechanism of generative AI, where the model 'samples' the next token from a predicted probability distribution." This sentence is cited with [@mcp-sampling], but the linked MCP Sampling documentation page (modelcontextprotocol.io/docs/concepts/sampling) makes no mention of token-level sampling or probability distributions. The source describes sampling only as a mechanism for servers to request LLM completions from clients. The etymology is an unsourced editorial interpretation presented as if it is documented in the spec.
No other issues found
- Architecture (Hosts/Clients/Servers), JSON-RPC 2.0, three capability types, Tools defined by JSON Schema, Sampling user-approval claims, and the November 2024 announcement date all checked against official sources and confirmed accurate.
- Streamable HTTP description is a minor simplification (SSE also used for streaming POST responses) but not materially false.
- Cursor adoption claim is weakly sourced but not contradicted by available evidence.
The etymology sentence was added in the latest revision (8909c8e8, rev summary: 'Clarified the etymology of Sampling in the context of LLM token generation'). While the LLM token-sampling connection is plausible general AI knowledge, it is not supported by the cited source and is presented inline as sourced. Recommended fix: either remove the sentence, source it to a reference that actually discusses it, or move it to an Analysis section with a clear editorial label. Human editor review recommended before any edit.