JevGuard MCP Server
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| TYPESAFE_API_KEY | No | API key for TypeSafe AI. Required for dispatching upstream when the cache misses; optional if only using local deterministic evaluation. |
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
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| jevguard_evaluateA | Executes the deterministic JevGuard evaluation pipeline including state pruning, closed-world escape injection, certainty calibration, and 0-token caching. |
| jevguard_calibrateA | Evaluates probability distributions across answers to identify ambiguity, low confidence (top_prob < 0.40), and flat distributions (dispersion_gap < 0.15). |
| jevguard_prune_stateB | Sanitizes and prunes complex JSON state payloads by removing nulls, empty collections, collapsing whitespace, and protecting against cyclic references. |
| jevguard_cache_fingerprintA | Calculates a canonical SHA-256 fingerprint from state and questions with volatile key masking (timestamp, trace_id, request_id) for 0-token deterministic caching. |
| evaluate_command_safetyB | Evaluates terminal/shell command safety for autonomous agents. Determines whether a command is destructive, requires human approval, or can execute autonomously using Noul, Score, and Choice certainty calibration. Returns policy: ALLOW_AUTONOMOUS, REQUIRE_HUMAN_APPROVAL, or DENY_DESTRUCTIVE. |
| verify_code_patchA | Evaluates whether a code diff or patch introduces security regressions, broken syntax, or critical system impact under a configurable risk tolerance (strict, balanced, permissive). |
| evaluate_decisionA | Evaluates architectural and implementation decisions with a simple list of options. Automatically injects closed-world neutral escape (UNRESOLVED_OR_OTHER) and calibrates probability dispersion. |
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 7 tools
Several tools occupy overlapping conceptual space: evaluate_decision, jevguard_calibrate, and jevguard_evaluate all involve evaluating, calibrating, or escaping closed-world options. Command safety and code patch verification are distinct, but an agent could easily misselect among the evaluation-related tools.
Naming is inconsistent: some tools use the jevguard_ prefix with snake_case (jevguard_prune_state, jevguard_calibrate), while others use bare action-style names (evaluate_command_safety, verify_code_patch). This mixed convention makes the tool set feel less predictable.
Seven tools is a reasonable size for a specialized evaluation server. The count is not excessive, though the overlapping evaluation/calibration tools could likely be consolidated without losing much functionality.
The tool set covers the core pipeline stages: state pruning, fingerprinting, safety evaluation, code patch verification, decision evaluation, calibration, and full pipeline execution. Minor gaps exist around direct inspection of cached fingerprints or explicit configuration of the full pipeline, but the core domain is covered.