Kryve Agent Evaluation MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Kryve Agent Evaluation MCP Generate a test plan for our customer support agent."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Kryve Agent Evaluation MCP
Un serveur MCP local et open source pour concevoir des tests d’agents IA, noter une exécution et documenter les preuves avant d’accorder davantage d’autonomie.
Il expose trois outils en lecture seule :
generate_test_plan: construit des cas nominaux, ambigus et sensibles selon le niveau de risque ;score_agent_run: calcule une auto-évaluation provisoire du résultat, des preuves déclarées, des validations et du coût des corrections ;evaluation_scorecard: renvoie les colonnes d’une scorecard réutilisable.
Installation
npm install
npm run buildPuis lance le serveur sur stdio :
npm startConfiguration d’un client MCP :
{
"mcpServers": {
"kryve-agent-eval": {
"command": "node",
"args": ["/chemin/vers/kryve-agent-eval-mcp/dist/src/index.js"]
}
}
}Related MCP server: AgentLens MCP Server
Tester avec MCP Inspector
npx @modelcontextprotocol/inspector node dist/src/index.jsPhilosophie
Une démo réussie ne prouve pas qu’un agent est fiable. Le test doit mesurer un résultat attendu, conserver des preuves, détecter les actions hors permission et rendre visible le coût des corrections humaines.
Pour comprendre le protocole et choisir une infrastructure :
Le kit CSV complémentaire est disponible dans le dépôt Kryve Agent Evaluation Kit.
Limites
Le score est une grille opérationnelle simple, pas une certification. Les champs sont déclarés par l’appelant : chaque résultat porte donc le statut unverified_self_report, n’est jamais éligible à une décision d’autonomie et doit être recoupé avec des preuves indépendantes.
La similarité textuelle sert uniquement d’indicateur provisoire. Elle ne remplace ni des critères structurés, ni un journal d’actions, ni une validation humaine vérifiable.
Licence
MIT — Kryve Solutions.
Available Tools
3 toolsevaluation_scorecardGet the Kryve evaluation scorecardARead-onlyIdempotent
Return a reusable scorecard structure for documenting AI-agent tests.
Inputs: no named parameters.
generate_test_planGenerate an agent test planARead-onlyIdempotent
Generate practical test cases and success criteria for an AI-agent use case.
Inputs: risk (required), useCase (required), tools.
score_agent_runScore an agent run provisionallyARead-onlyIdempotent
Create an unverified self-assessment of outcome, evidence, control and correction cost. Never use it alone to grant autonomy.
Inputs: task (required), actualOutcome (required), expectedOutcome (required), evidenceProvided (required), +4 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 3 tools
Each tool addresses a distinct phase of evaluation: generating test plans, scoring an agent run, and retrieving a scorecard structure. There is no functional overlap or ambiguity between them.
Two tools follow a clear verb_noun pattern (generate_test_plan, score_agent_run), but evaluation_scorecard is noun-led rather than verb-led. The inconsistency is minor and all names remain readable and predictable.
Three tools is an appropriate, focused scope for an evaluation-oriented MCP server. Each tool serves a core need without bloat, making the surface easy to navigate.
The set covers planning, assessment, and structure retrieval, but lacks an explicit verification or finalization tool. The 'unverified' caveat on score_agent_run hints at this gap, but the core workflow is still functional.
Maintenance
Related MCP Connectors
Run AI evaluations, record agent traces and gate releases on evidence, with scoped org access.
Run AI agent evaluations on your own model keys, and publish runs others can review and re-run.
Test the voice agents you run: scored transcripts, pass/fail verdicts, latency and WER metrics.
LLM evaluation: compare models on cost and quality, score outputs, run benchmarks, analyse agents.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceEnables LLM evaluation and observability by uploading documents, building test sets, running RAG pipelines, and automatically scoring answers for groundedness, hallucination risk, retrieval quality, latency, and cost, with tools exposed to MCP-compatible clients.1MIT
- AlicenseNot gradedqualityCmaintenanceEnables AI agents to access observability and evaluation data, including run history, span traces, LLM-as-judge evaluation results, and regression reports.MIT
- AlicenseAqualityBmaintenanceProvides coding agents with visibility into test health through tools for flaky test detection, test quality linting, and LLM evaluation harness, enabling them to triage failures, review test quality, and check prompt changes for regressions.9Apache 2.0
- AlicenseNot gradedqualityBmaintenanceEnables running and grading AI agent evaluations, and applying runtime guardrails that check or block unsafe tool calls through dedicated guardrail tools.MIT