Cohere's ATE Dataset: Only 2.6% of 690K MCP Tools Can Independently Complete Job Tasks

Cohere Labs released the Agentic Task Ecosystem (ATE) dataset, cataloging nearly 700K tools scraped from public MCP servers, offering a window into what developers are actually building AI agents to do. The headline finding: in tests of whether a single tool can independently complete an occupational task, only 2.6% passed—suggesting that despite the ecosystem's size, very few tools can autonomously support real job-task automation.

Confirmed

• ATE was released by Cohere Labs and contains over 690K tools from public MCP servers
• The dataset is positioned as a resource for evaluating whether agents can autonomously complete real occupational tasks and for examining automation trends in the agent ecosystem
• In the "can a tool independently complete an occupational task" test, the pass rate was only 2.6%
• The Cohere team argues that AI's impact on the future of work depends not only on model capability but also on the actual coverage of the tool ecosystem

Why it matters

• The dataset offers a rare, large-scale first-hand window into the MCP tool ecosystem and agent automation trends
• The 2.6% pass rate quantifies the huge gap between today's agent ecosystem and "replacing occupational tasks," calibrating the narrative that AI is on the verge of fully automating work

2026-09-03 ~ 2026-09-04 · 5 related posts

• [source] Cohere Labs releases ATE dataset of ~700K tools from public MCP servers — Cohere_Labs · 2026-09-03
• Cohere Labs tests 690K+ tools for agents — only 2.6% can complete a job task alone — cohere · 2026-09-04

3 near-duplicate retellings: Cohere_Labs · Cohere_Labs · cohere