LLM Eval Harness
Run a golden set and gate LLM model swaps on accuracy, with latency and cost per request.
MCP server · by razilzul
About
LLM Eval Harness is an MCP server published by razilzul. Marketplaces we read that list it: 1 (Official MCP Registry). We first saw it on 2026-10-18 and have recorded no change since. As of our marketplace pass on 2026-10-22.
What it asks for
- Runs on your machine
- Secrets: 2
- Settings: 2
Requires: ANTHROPIC_API_KEY (secret), GEMINI_API_KEY (secret)
What it declares, asks for and runs, in full
Only install plugins from developers you trust. We list what marketplaces publish and do not test plugins.
Works with
- Claude Code
- Codex
- Cursor
- VS Code
- Copilot CLI
From the marketplaces that list it and the MCP transport it declares, as each client documents them. We do not test plugins.
Connect
Commands and configs built only from what its marketplaces declare. We do not run them for you and do not test plugins.
As a local MCP server: Docker image
Built from what Official MCP Registry declares.
claude mcp add --env "ANTHROPIC_API_KEY=<YOUR_ANTHROPIC_API_KEY>" --env "GEMINI_API_KEY=<YOUR_GEMINI_API_KEY>" --transport stdio llm-eval-harness -- docker run -i --rm -e ANTHROPIC_API_KEY -e GEMINI_API_KEY ghcr.io/razilzul/llm-eval-harness:1.0.0[mcp_servers.llm-eval-harness]
command = "docker"
args = ["run", "-i", "--rm", "-e", "ANTHROPIC_API_KEY", "-e", "GEMINI_API_KEY", "ghcr.io/razilzul/llm-eval-harness:1.0.0"]
[mcp_servers.llm-eval-harness.env]
ANTHROPIC_API_KEY = "<YOUR_ANTHROPIC_API_KEY>"
GEMINI_API_KEY = "<YOUR_GEMINI_API_KEY>"{
"mcpServers": {
"llm-eval-harness": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"ANTHROPIC_API_KEY",
"-e",
"GEMINI_API_KEY",
"ghcr.io/razilzul/llm-eval-harness:1.0.0"
],
"env": {
"ANTHROPIC_API_KEY": "<YOUR_ANTHROPIC_API_KEY>",
"GEMINI_API_KEY": "<YOUR_GEMINI_API_KEY>"
}
}
}
}docker run -i --rm -e "ANTHROPIC_API_KEY=<YOUR_ANTHROPIC_API_KEY>" -e "GEMINI_API_KEY=<YOUR_GEMINI_API_KEY>" ghcr.io/razilzul/llm-eval-harness:1.0.0Replace the values in angle brackets with your own. We show the names of the settings, never values.
Where it is published
| Marketplace | Status | Version | Figures | Verified | First seen |
|---|---|---|---|---|---|
| Official MCP Registry (Open the registry record) | Listed | 1.0.0 | Forks (GitHub): 0 Stars (GitHub): 0 |
Not marked |
Each figure is the marketplace's own and is never added up across marketplaces. A marketplace that publishes no figure shows none: no number is not zero.
What changed
- Became a product we track
- Official MCP Registry: appeared
Similar plugins
Products whose own names, descriptions, tools and skills are closest in meaning to this one, on every marketplace we read. Closeness is ranked within this list only.
-
Read, measure and explain LLM eval runs of a digline suite. Cannot promote a baseline.
-
Benchmark your own prompt on 100+ LLMs with real API calls; pick the cheapest model that passes.
-
llmtest-mcp io.github.tjacquesson/llmtest-mcp
Benchmark AI models on real prompts. Find cheaper, faster alternatives across 340+ models.
-
Compare text LLMs across OpenRouter, Bedrock, Vertex AI and Foundry with automated grading.
-
Decide, design, gate, judge, and defend evaluations for LLM, agent, and retrieval systems. Not ordinary unit tests.
-
Configures and runs LLM evaluation using Promptfoo framework. Use when setting up prompt testing, creating evaluation configs (promptfooconfig.yaml), writing Python custom…
-
metrillm io.github.MetriLLM/metrillm
Benchmark local LLM models — speed, quality & hardware fitness verdict from any MCP client
-
Read and score Opik traces, experiments, datasets and prompts from your AI assistant.
-
llm-observability-orchestration-agent-langchain-community ai.getvda/llm-observability-orchestration-agent-langchain-community
LLM Observability & Orchestration Agent (LangChain Community)
-
LLM serving benchmarks, SGLang model Day-0 support, profiling, code review, prod incident triage, PR-history dossiers, and SGLang explainer videos.
Similar plugins by marketplace
The same question asked of one marketplace at a time: is there an analog of this plugin there?
Has a listing on Claude Code: claude-plugins-community (5)
-
AI agent evaluation sandbox for Claude Code. Holdout scenario testing with Doer/Judge/Adversary/Observer roles, probabilistic satisfaction scoring, and append-only JSONL audit…
-
Synthetic monitoring for LLM inference endpoints. Probe TTFT, latency, throughput, and errors across OpenAI, Anthropic, Google, Azure, Bedrock, and any OpenAI-compatible server.
-
Multi-LLM orchestration MCP server. Delegate tasks to Claude Code, Codex, and Gemini CLIs with session management, retry logic, async job orchestration, and token optimization.
-
Claude model selection advisor — recommends optimal model for each task via shell hook. Zero attack surface.
-
Skills for working with Langfuse, the open-source LLM engineering platform for tracing, prompt management, and evaluation.
Has a listing on Claude Code: claude-plugins-official (5)
-
Skills for adding DeepEval evaluations, tracing, datasets, Confident AI reports, and iterative improvement loops to AI applications.
-
Skills for working with Langfuse, the open-source LLM engineering platform for tracing, prompt management, and evaluation.
-
Skills for tracing, evaluating, and improving AI agents with MLflow. Supports the full agent improvement loop: instrument → trace → evaluate → iterate → validate.
-
OpenTelemetry observability for Claude Code sessions. Captures tool calls, LLM invocations, token usage, and errors as OTel traces. Send telemetry to Dash0 or any…
-
DataRobot skills for AI/ML workflows — model training, deployment, predictions, feature engineering, monitoring, explainability, data preparation, App Framework CI/CD, and…
Has a listing on Claude: knowledge-work-plugins (5)
-
Skills for working with Langfuse, the open-source LLM engineering platform for tracing, prompt management, and evaluation.
-
DataRobot skills for AI/ML workflows — model training, deployment, predictions, feature engineering, monitoring, explainability, data preparation, App Framework CI/CD, and…
-
AI-first project auditor and re-engineer based on the 9 design principles and 7 design patterns from the TechWolf AI-First Bootcamp
-
Build multimodal AI applications with Pixeltable -- tables, computed columns, embedding search, UDFs, tool-calling agents, and 25+ AI provider integrations.
-
Agent skills for Qdrant vector search covering scaling, performance optimization, search quality, monitoring, deployment, model migration, version upgrades, and SDK usage across…
Has a listing on Codex: openai-curated (openai/plugins) (5)
-
Evaluate Codex skills and plugins from chat with a beginner-friendly start command, local-first reports, token budget explanations, and guided benchmarking.
-
Skills for NVIDIAs ecosystem spans GPU acceleration, CUDA, AI agents, inference, robotics, Physical AI, Omniverse, and simulation. This plugin helps you understand the pieces…
-
Build with OpenAI APIs, Agents SDK, and ChatGPT Apps, and create and save OpenAI API keys from Codex.
-
Available for US1 customers only. Analyze, investigate, and act on your Datadog telemetry directly from ChatGPT using natural language. Ask questions about your production…
-
CrowdStrike Falcon Foundry development skills for building cybersecurity applications on the Falcon platform. Includes UI development, collections, functions, workflows, API…
Has a listing on Community GitHub marketplaces (5)
-
Decide, design, gate, judge, and defend evaluations for LLM, agent, and retrieval systems. Not ordinary unit tests.
-
Configures and runs LLM evaluation using Promptfoo framework. Use when setting up prompt testing, creating evaluation configs (promptfooconfig.yaml), writing Python custom…
-
LLM serving benchmarks, SGLang model Day-0 support, profiling, code review, prod incident triage, PR-history dossiers, and SGLang explainer videos.
-
Skills for adding DeepEval evaluations, tracing, datasets, Confident AI reports, and iterative improvement loops to AI applications.
-
Scaffolding, running, documenting, and publishing AI evaluations. Ships skills and commands for setting up eval workspaces, creating custom evals (or adapting existing…
Has a listing on Copilot CLI: github/awesome-copilot (5)
-
Run a lightweight, role-separated AI development team with flexible tools, developer-selected models, proportionate planning, and optional QA.
-
See your local Copilot token usage by day, model, and session, and get tips to use it more efficiently.
-
Three-layer verification pipeline for AI output. Extracts claims, finds sources, and flags hallucination risks so humans can verify before acting.
-
Arize AX platform skills for LLM observability, evaluation, and optimization. Includes trace export, instrumentation, datasets, experiments, evaluators, AI provider integrations…
-
Visualizes live session AI-credit usage as a Pac-Man board with pellets, ghosts, fruit milestones, and game-over limits.
Has a listing on Copilot CLI: github/copilot-plugins (5)
-
Toolkit for building Microsoft 365 Copilot declarative agents — scaffolding, JSON manifest development, and capability configuration.
-
Build and manage Microsoft Dataverse solutions using natural language. Includes table/column creation, solution lifecycle, data operations, and MCP server configuration.
-
model-apps
Build and deploy Power Apps generative pages for model-driven apps with specialist agents for planning, entity creation, and parallel code generation. Requires PAC CLI >= 2.7.0…
-
workiq
Query Microsoft 365 data with natural language — emails, meetings, documents, Teams messages, and more.
-
workiq-productivity
WorkIQ productivity skills powered by the local WorkIQ CLI — read-only insights across email, meetings, calendar, Teams, SharePoint, projects, and people.
Has a listing on Cursor: cursor/plugins (5)
-
Fan large tasks out across parallel cloud agents with planners, workers, verifiers, and structured handoffs.
-
Iterative self-referential AI loops using the Ralph Wiggum technique.
-
Inspect and manage feature gates, experiments, dynamic configs, and metrics.
-
Incremental transcript-driven memory updates for AGENTS.md using high-signal bullet points only.
-
Build apps, scripts, and automations with the TypeScript SDK.
Has a listing on Docker MCP Catalog (5)
-
Read and score Opik traces, experiments, datasets and prompts from your AI assistant.
-
Bridges multiple AI models and CLIs, enabling orchestrated workflows across Claude Code, Gemini CLI, Codex CLI, and other AI development tools.
-
Wayfound’s MCP server allows business users to govern, supervise, and improve AI Agents.
-
Analyze documents, images, videos, and audio.
-
The Official Model Context Protocol (MCP) server for Kagi Search & other tools.
Has a listing on Official MCP Registry (5)
-
Read, measure and explain LLM eval runs of a digline suite. Cannot promote a baseline.
-
Benchmark your own prompt on 100+ LLMs with real API calls; pick the cheapest model that passes.
-
llmtest-mcp io.github.tjacquesson/llmtest-mcp
Benchmark AI models on real prompts. Find cheaper, faster alternatives across 340+ models.
-
Compare text LLMs across OpenRouter, Bedrock, Vertex AI and Foundry with automated grading.
-
metrillm io.github.MetriLLM/metrillm
Benchmark local LLM models — speed, quality & hardware fitness verdict from any MCP client
Has a listing on npm registry (MCP servers) (5)
-
One workspace for every AI coding assistant. Shared tasks, memory, and governance.
-
Read-only MCP server exposing the dsh-plugin-certification registry for DeepSeek Harness plugins.
-
easytable-mcp io.github.chrischall/easytable-mcp
easyTable restaurant reservations for Claude — availability, and create/modify/cancel bookings
-
Human-in-the-loop for AI agents over MCP elicitation: confirm/form dialogs plus a capability doctor
-
Browser-local and CLI static evidence for deployed AI model artifacts.
Found by meaning, not by shared code or a shared repository: similar plugins stay separate products, and this is not a review or a recommendation.
Want this mark on your plugin in our catalog?
Prove you maintain it with a file in its repository, a file on your domain or a DNS record, and you can show this mark on the card, with live banners for your site and README. How verification works
You also get alerts when a marketplace lists it, drops it, changes its verified mark or what it asks for, and its remote endpoint goes on monitoring as a web service.
Data comes from the Official MCP Registry (registry.modelcontextprotocol.io), CC0 1.0. Figures come from our own regular passes over the marketplaces, last read on . We publish no contact details for plugin publishers.