# MCPMark-Verified ![Category](https://img.shields.io/badge/Category-MCP%20Tool--Use%20Agent-1f6feb) ![Metric](https://img.shields.io/badge/Metric-%25%20Success-2da44e) ![Saturation](https://img.shields.io/badge/Saturation-Low-d29922) ![Created](https://img.shields.io/badge/Created-2026-555) > **Quick answer:** MCPMark-Verified is a benchmark testing AI agents on tasks that require correctly interacting with Model Context Protocol (MCP) servers and tools — connecting to external systems, invoking MCP tool calls, and completing verified end-to-end workflows. Kimi K3 scores 94.5% as of July 2026. ## At a Glance **What it tests:** An agent's ability to correctly discover, connect to, and invoke tools exposed via the Model Context Protocol (MCP) to complete real tasks. **Why it matters:** MCP has become a standard way for AI agents to connect to external tools and data sources. MCPMark-Verified measures how reliably a model can operate within this now-widespread protocol. **Known limitations:** As an emerging benchmark, the exact MCP server/tool inventory and task composition are not yet independently published outside its citation by Moonshot AI. ## What MCPMark-Verified Measures MCPMark-Verified evaluates an agent's ability to interact correctly with tools and data sources exposed through the Model Context Protocol (MCP) — an increasingly standard interface for connecting LLM agents to external systems. Tasks require correct tool discovery, parameter construction, and multi-step MCP tool invocation, with outcomes verified against expected results. ## Benchmark Specifications | Field | Value | |---|---| | Task category | Agent / MCP tool-use | | Metric | % task success | | Saturation | Low | | Created by | Not yet independently documented | ## How MCPMark-Verified Is Scored Agents complete tasks requiring MCP tool interaction, with outcomes verified against expected results, producing an aggregate % task success rate. ## State-of-the-Art Results | Rank | Model | Score | Source | Date | |---|---|---|---|---| | 1 | [Kimi K3](/models/moonshot-ai/kimi-k3) | 94.5% | [Kimi K3 technical report](https://www.kimi.com/blog/kimi-k3) | 2026-07 | *Score sourced from Moonshot AI's Kimi K3 announcement, July 2026.* ## MCPMark-Verified on Benchgen > No Benchgen results yet — [be the first to run MCPMark-Verified](/benchmarks/wu-et-al/mcpmark-verified). ## MCPMark-Verified vs Other Benchmarks | Benchmark | What it tests | Saturation | |---|---|---| | **MCPMark-Verified** | MCP protocol tool-use correctness | Low | | [MCP Atlas](/benchmarks/scale-ai/mcp-atlas) | MCP tool-use agentic workflows | Low | | [Toolathlon-Verified](/benchmarks/hkust-nlp/toolathlon-verified) | Multi-tool, multi-step agentic workflows | Low | | [BFCL v3](/benchmarks/berkeley/bfcl-v3) | Function-calling accuracy | Medium | ## Run MCPMark-Verified on Your Model Benchgen lets you run MCPMark-Verified against your own agent's MCP integration, tracking tool-use task success rates over time. ## Frequently Asked Questions

What is MCPMark-Verified?

MCPMark-Verified is a benchmark testing AI agents on Model Context Protocol (MCP) tool-use tasks, with verified task setups and success criteria.

What does a good score look like on MCPMark-Verified?

Kimi K3 reports 94.5% as of July 2026, a very strong result reflecting reliable MCP tool-use capability.

Who created MCPMark-Verified?

MCPMark-Verified's originating team is not yet independently documented outside of its citation in Kimi K3's July 2026 technical report.

Rank	Model	Score
1	kimi-k3	94.5
2	k2-horizon-375b-a23b	67.7
3	spark-x2-5-4b	14.2
4	spark-x2-5-4b	14.2
5	spark-x2-5-1-7b	2.3
6	spark-x2-5-1-7b	2.3

MCPMark-Verified — Results

MCPMark-Verified — Evaluations