jev-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| JEV_MCP_MOCK | No | Set to '1' to use local deterministic judge (tests/demos only) | |
| JEV_MCP_MODEL | No | Model to use, default 'jev-latest' | jev-latest |
| JEV_MCP_BLOCK_AT | No | Block threshold (screen), default 0.75 | 0.75 |
| TYPESAFE_API_KEY | No | Live TypeSafe API key | |
| JEV_MCP_REVIEW_AT | No | Review threshold, default 0.5 | 0.5 |
| TYPESAFE_BASE_URL | No | Optional API root | |
| JEV_MCP_AUTO_ACCEPT | No | Auto-accept threshold, default 0.8 | 0.8 |
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": true
} |
| resources | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| jev_evaluateA | Escape hatch: send shared state plus named noul/choice/score questions to TypeSafe Jev. Use when no other jev_* recipe fits. Jev does not write code or prose. Questions in one call run in parallel. Returns typed answers, probabilities, confidence, usage, and action auto|review|escalate. |
| jev_coding_loopA | Call before spending a frontier turn on retry/stop/model-tier. One Jev fan-out returns next (continue|retry|ask_user|stop), model_tier (cheap|standard|reasoning), risk, focus, and noul flags done_enough / needs_more_context / tests_likely_fail. Policy in code maps confidence to action auto|review|escalate. Does not edit files. |
| jev_reviewA | Score a proposed diff against the request: correctness, spec-match, test-gap, blast-radius, plus noul safe_to_apply. Composite weights live in code. Call before declaring a fix done. Does not apply the patch. |
| jev_verifyA | Check each claim against provided evidence (PR description, agent brief, docs, diffs). Returns per claim: verified|contradicted|unsupported, probabilities, confidence, and auto vs review. Prefer this over asking a chat model to 'double-check'. |
| jev_screenA | Judge fetched or pasted text before the agent reads it: prompt-injection probability, substance, and optional relevance to purpose. Recommendation: pass|review|block|skip. Use on untrusted web pages, issues, and pastes. Not for first-party repo files. |
| jev_rankA | Rank files, symbols, errors, or skills against a plain-language query. No embeddings. One Choice over candidate ids plus a Noul that the top hit actually answers the query (so a forced winner cannot masquerade as a match). Max 250 candidates per Jev call; larger lists are chunked then re-ranked. Pass candidates in; this server does not index the repo. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| pack-coding-loop | Exact question JSON used by the coding-loop recipe. Tune thresholds in code, not by rewriting Jev into a chat prompt. |
| pack-review | Exact question JSON used by the review recipe. Tune thresholds in code, not by rewriting Jev into a chat prompt. |
| pack-verify | Exact question JSON used by the verify recipe. Tune thresholds in code, not by rewriting Jev into a chat prompt. |
| pack-screen | Exact question JSON used by the screen recipe. Tune thresholds in code, not by rewriting Jev into a chat prompt. |
| pack-rank | Exact question JSON used by the rank recipe. Tune thresholds in code, not by rewriting Jev into a chat prompt. |
TDQS
Scored across 6 tools
Each recipe targets a distinct decision point (loop control, diff review, claim verification, text screening, candidate ranking), and `jev_evaluate` is explicitly scoped as an escape hatch rather than a competing operation. There is only a mild risk that agents reach for the generic evaluate tool instead of the specialized recipes.
All tools share the lowercase `jev_` prefix and use underscores, which gives the set a clear visual pattern. However, the names mix imperative verbs (`review`, `verify`, `screen`, `rank`, `evaluate`) with one noun-style name (`coding_loop`), a minor consistency deviation.
Six tools is a well-scoped size for a decision-support server. Each tool fills a distinct role with no apparent bloat or redundant utility.
The tool surface covers the major agent workflow decision points: whether to continue, whether to apply a diff, whether claims hold up, whether text is safe to read, and which candidates best match a query. The generic `jev_evaluate` fallback plus usage/action outputs prevents obvious dead ends.