Exact structural diff plus meaning-change analysis protocol - what actually changed between two versions, and why it matters. Free MCP plugin for Claude and ChatGPT.
Two well-named tools with strong, LLM-optimized descriptions (200+ chars each). Input schemas are present with proper types and constraints (enums, minLength). However, output schemas are not documented, the protocol response is free-form text rather than structured JSON. Error handling is implicit (relies on the LLM to interpret the analysis). Tool composition is sound: semantic_diff and diff_only are cleanly separated concerns. Naming follows verb_noun pattern (semantic_diff, diff_only). Parameter descriptions are clear and include constraints (minLength=1, enum values). Missing: explicit output schema documentation, error recovery guidance, and per-request _meta logLevel support.
Return only the exact computed line diff between two texts, with true added/removed counts and no analysis. Use when the raw diff is wanted rather than an interpretation.
Compare two versions of a document, contract, policy or code and explain what changed IN MEANING — not just which lines moved. The diff is computed exactly (models miscount changes by eye), then analysed for substance: rights gained or lost, liability shifted, protections removed, behaviour altered. Use whenever two versions are compared, or the user asks 'what changed', 'what's different', or 'what did they sneak in'.
Output schema not documented. Both tools return free-form text (protocol analysis or diff string) rather than structured JSON. LLMs cannot plan downstream operations or extract specific fields. Responses should be structured objects with typed fields (e.g., {verdict, critical_changes[], notable_changes[], removed_protections[], noise, cannot_tell}).
No error handling guidance. If inputs exceed MAX_LINES (400) or MAX_CHARS (60000), the tool silently truncates or fails. Error responses should tell the LLM what to do: 'Input exceeds 60,000 characters. Summarize the before/after versions and retry, or split into smaller diffs.'
Missing per-request _meta logLevel support. The server does not expose per-request logging configuration as per current MCP spec (2026-07-28). Clients cannot control verbosity on a per-call basis.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | 2026-07-28+ | v2 |
No idempotency or confirmation pattern. Both tools are read-only (no state mutation), but the semantic_diff tool produces a lengthy analysis that could be expensive to recompute. No caching or idempotent hint is documented.
Lens parameter enum is not validated server-side. If an LLM passes an invalid lens value, the tool should return a clear error listing valid options (legal, privacy, code, policy, prose, auto) rather than silently defaulting to 'auto'.