Rerun Launches MCP Server for AI Agent-Powered Robotics Data Inspection
Key Takeaways
- ▸Rerun's viewer-mcp MCP server lets Claude, Codex, and other agents interact directly with the Rerun Viewer for robotics data inspection and debugging
- ▸Agents can autonomously debug logging scripts, create custom blueprints with multiple views, explore recordings, and capture screenshots through the visual interface
- ▸Headless rendering support enables the MCP to work in CI/CD pipelines and background server environments without a display, unlocking automated validation workflows
Summary
Rerun has announced the availability of viewer-mcp, a Model Context Protocol (MCP) server that enables AI agents such as Claude and Codex to directly interact with the Rerun Viewer for inspecting and debugging robotics and embodied AI data. The MCP server allows agents to autonomously perform tasks like debugging logging scripts, creating custom visualization blueprints, exploring recordings, adjusting viewer settings, and capturing screenshots—treating the visual interface like any other tool.
The viewer-mcp operates via gRPC to connect to a running Viewer instance, making it straightforward to integrate with existing MCP clients. Rerun provides quick-start configurations for Claude Code and Codex, as well as generic mcp.json setup for other MCP clients. Notably, the server supports headless rendering, enabling agents to operate in CI pipelines and background server environments without requiring a display, opening new workflows for automated data inspection and validation in production systems.
The capability addresses a gap in agentic AI workflows: while agents can interact with APIs and code, they often lack direct access to visual tools designed for human inspection. By exposing the Rerun Viewer through MCP, Rerun enables agents to leverage its specialized robotics visualization capabilities programmatically, making it easier to debug complex sensor data, validate data collection scripts, and explore large datasets at scale.
- Easy integration: Claude Code and Codex users can add the MCP with a single command; other clients can configure via standard mcp.json
Editorial Opinion
This is a smart tactical move that extends Rerun's value into the agentic AI era. Rather than building agents into their product, Rerun is meeting agents where they are—via MCP—which sidesteps the complexity of maintaining multiple AI integrations while giving Claude and Codex users immediate access to specialized robotics visualization. The headless rendering capability is particularly valuable; it shifts robotics data inspection from a manual, GUI-bound task to an automated process that fits into modern CI/CD workflows and agentic reasoning loops. If agents can autonomously reason about visual sensor data, it could accelerate debugging and validation cycles in robotics development.
