You’ve been using custom MCP servers with qbash for months, and asking us for an official one. It’s here.
The official qbash MCP is live
- Build Tasks. Describe a job and let your agent build a hosted Python Task.
- Run Tasks. Call existing Tasks from your AI client and read the results.
- Share Markdown Memories. Give Claude Code, Codex, and your team the same notes across sessions.
- Access your connectors. Use your connected apps from your AI client, with account access you approve.
Connect your AI tool
Add https://mcp.qbash.com/mcp to your MCP client and sign in with qbash. The help docs have setup instructions for ChatGPT, Claude, Claude Code, Codex, and other MCP clients.
Read the official MCP server docs →
Two years into our AI platform adventure, we’ve watched people switch tools, adopt new ones, and hand more work to agents. Throughout that, qbash Tasks have become functions that companies rely on to keep their operations running.
As tools and agent teams proliferate, those shared functions become more valuable. qbash keeps these critical functions in one managed layer, so people can call them from the tools they use, and systems and agents have a control layer for the work they do. Our strategy deck below shares more of how we’re thinking about this.
From the blog
The AI Ticker, October 22 to 29
Google announced Gemini 4 Carbon. Initial access is going to selected cyber defenders and trusted partners, with broader availability still to come. Google reports gains in coding, long-running agent work, and vulnerability detection.
Anthropic filed for an IPO. Its October 26 announcement names the proposed Nasdaq symbol ANT. The company has not yet determined the offering’s share count or price range.
OpenAI announced AI-generated results in mathematics, computer science, and physics. It is publishing proof artifacts and working with external researchers to examine and explain the results, including a claimed refutation of the Exponential Time Hypothesis.
Agent Skills is moving to the Agentic AI Foundation. Anthropic is donating the standard, name, and associated infrastructure to the Linux Foundation project. More than 70 agent platforms have integrated the format, according to Anthropic.
AI’s drug-development work includes a lot of paperwork. In Anthropic’s latest update, Enveda’s test on 1,061 pages reduced human document-review time from over 25 hours to under three. The wider drug research remains mostly preclinical.
Google’s Genie 4 adds sound to interactive worlds. Its new research model generates audio alongside scenes you can steer with natural-language instructions. Access has started with trusted testers; Google AI Ultra and early API access are still to come.
Claude’s Google Calendar connector is working again. Anthropic says failures affecting calendar search and event listing ran from October 26 to 28 across claude.ai, Claude Code, and Cowork. The incident is now resolved.
A coding-agent game gets four-player multiplayer. Simon Willison’s Battle Drones update adds a new map, four-player networking, and share links or QR codes to join. His write-up includes the prompt used to have Claude make the change.