Google gives AI agents fresher docs with Gemini API Docs MCP and Agent Skills

Google published two new tools on April 1 for developers building coding agents on Gemini: Gemini API Docs MCP and Agent Skills. The goal is straightforward: give agents access to current documentation and better working patterns so they stop generating code against outdated APIs.

Gemini API Docs MCP connects an agent to live Gemini documentation, SDKs, and model information through the Model Context Protocol. Agent Skills adds best-practice instructions, resource links, and implementation patterns meant to steer the agent toward the right SDK usage.

Google says the combination reached a 96.3 percent pass rate in its internal evals while using 63 percent fewer tokens per correct answer than vanilla prompting. That makes this a practical launch for teams building agentic developer tools, not just another model-side tweak.

For CIOs and product teams, the bigger point is that agent tooling becomes more reliable when it can pull real-time context directly from the vendor. That can reduce errors, lower maintenance overhead, and make coding assistants more viable in production workflows.