Runheed news
New things you can do with Runheed, one post per change. Follow along with the RSS feed. Every smaller change, dated, is in the changelog.
Error reports start at the line in your code
When a library wraps your error in its own, the Python and Node.js SDKs now put the frames from your code first, so your agent opens the right file.
Every Runheed plan starts with a 7-day trial
Your workspace's first plan now starts with a 7-day trial of the full plan, so errors arrive and your agent reads them from day one.
Your agent knows what to fix first
Project health now names the busiest unresolved error and any error that came back after a fix, and alerts say where each error happened.
Errors arrive with what caused them
The Node and Python SDKs now send the chain of underlying errors, so a vague TypeError: fetch failed arrives with the real reason, such as ENOTFOUND.
See what your agent sees, with no account
The demo at runheed.com/demo shows one sample production error, the MCP calls a coding agent makes to read it, and a short Claude Code session that fixes it.
Connect your agent with just the URL
The Runheed MCP server now signs in over OAuth, so claude.ai, ChatGPT, Claude Code, Cursor and Codex connect with the server URL and no key to paste.