answer-check-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| ANSWER_CHECK_UK_INDEX | No | Path to the legal-rag-router index (about 360 MB) used to enable resolve_uk_citation. Can also be set with --uk-index. Default: none. | |
| ANSWER_CHECK_JUDGE_URL | No | OpenAI-compatible judge endpoint URL. When set, judge mode is enabled. Default: none (judge mode off). | |
| ANSWER_CHECK_JUDGE_MODEL | No | Model name for a custom, unmeasured judge. Mutually exclusive with ANSWER_CHECK_JUDGE_PRESET and requires ANSWER_CHECK_JUDGE_URL. Default: none. | |
| ANSWER_CHECK_JUDGE_PRESET | No | Measured judge preset to use. Mutually exclusive with ANSWER_CHECK_JUDGE_MODEL and requires ANSWER_CHECK_JUDGE_URL. Default: qwen27-ar4 when only ANSWER_CHECK_JUDGE_URL is given. | qwen27-ar4 |
| ANSWER_CHECK_JUDGE_TIMEOUT | No | Judge request timeout in seconds. | 30 |
| ANSWER_CHECK_JUDGE_MAX_CALLS | No | Maximum number of judge calls per check. | 40 |
| ANSWER_CHECK_JUDGE_API_KEY_ENV | No | Name of the environment variable that holds the judge API key. The key itself is never passed as a flag. Default: none. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| check_answerA | Check each sentence of an answer against the sources it should rest on. Each sentence comes back 'supported', 'unsupported' (with the claims no source backs and, where possible, the source values to use instead) or 'unchecked'. Figures stated in the question count as supported. No model is called, and the same input always gives the same result. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 1 tool
There is only one tool, so there is no possibility of confusing it with another. Its purpose—sentence-level verification of an answer against sources—is stated precisely and includes clear output semantics (supported/unsupported/unchecked).
The single name check_answer follows a clean verb_noun snake_case convention. With one tool there is no opportunity for inconsistent naming to creep in.
A one-tool surface is thin, but the server appears deliberately scoped to a single deterministic verification operation, so the tool is not arbitrary. It sits at the borderline of the count guideline rather than being a clear mismatch.
The tool covers its stated domain: it checks each sentence, reports per-sentence status, and supplies replacement values where possible. Minor gaps exist—no batch/multi-answer mode or configuration of source handling—but core verification is covered with no dead ends.