What it does
The tool reviews one answer per evaluation against a rubric that the caller supplies: required concepts, accepted synonyms, optional numeric requirements and the answer text. It includes no question bank or answer corpus. A concept score is the share of required concepts matched, capped at 1.0, and a numeric failure halves the score. Results are labelled correct, partial or wrong, and include matched and missing concepts, numeric check details and a review flag.
It suits assistive practice, where an AI agent or developer wants structured feedback on a written answer. It is not an official certification grader. Keyword matching can miss correct paraphrases and can accept misleading surface matches, so the rubric and every result should be reviewed. Agents can call the single stdio tool evaluate_answer, or send requests to a REST endpoint.
The REST server runs on Python 3.10 or later with the standard library. The MCP adapter needs the official Python SDK, installed from the committed lockfile. By default the service listens on localhost. Local single-key mode requires an API key of at least 24 characters, and deployments use a separate hashed key per client. The stdio transport needs no internet endpoint or API key. The MCP server keeps local daily usage counts for 90 days, which can be printed or disabled. Remote telemetry is off by default and, if a user opts in, sends only a random installation ID, package version and outcome.
Limitations and requirements
Requires Python 3.10 or later and the official Python SDK for the MCP adapter. REST use requires API keys: at least 24 characters in local mode, and a distinct hashed key per client when deployed. Rate limit of 60 requests per minute per client ID, held in memory and reset on restart. Request bodies are capped at 64 KiB. Keyword matching can miss correct paraphrases and accept misleading matches; all results require human review. Remote telemetry is opt-in and not included in PyPI version 0.1.1; it appears in 0.1.2 and collects only after explicit enablement.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
Technical Answer Validator
Rubric-based technical answer feedback for AI agents; human review required.
v0.1.2As shown on registry.modelcontextprotocol.io when we last read it