SpeedrunCI connects to any app that supports MCP, the open standard AI apps use to talk to other tools. There's nothing to install and no API key to copy.
SpeedrunCI knows your builds: what's red, why, since when, and what changed since the last green build. Your coding agent knows your code. Together they close the loop.
“Are we green?” Your AI checks the pipelines you can see and tells you what's failing, and since when.
It reads the failure the way a build engineer would: the first real error, the log around it, the commits since the last green build, and whether the same failure shows up elsewhere.
In Claude Code, Codex, Cursor or VS Code, the agent edits your local checkout and the fix gets pushed from there. SpeedrunCI never writes to your repository.
Once you approve, the AI starts the rebuild, follows the build until it finishes, and tells you when it's green.
In Claude or ChatGPT you get the same diagnosis, plus a clear description of the change that would fix it, ready to hand to whoever owns the code.
You don't need to read a build log to know what broke. Ask in your own words and get an answer you can act on.
“Why is main red? Fix it, push, rebuild, and tell me when it's green.” The whole loop, without leaving your editor.
“Did my material break the cook? Explain it like I'm not a programmer.” A plain-language answer, without the stack trace.
“What changed between build 412 and 418?” “Write a bug report with the build details.” The facts, ready to paste into your tracker.
“Summarize this week's build health for standup.” Where things stand, in a paragraph you can read out.
“Which pipelines are failing, and what was their last green build? Show me the log around the first error.” Triage done before you open the dashboard.
“Are all our projects green right now?” One question, every project you can see.
Read everything about your builds, and act on them with your approval. More toolsets are on the way.
See what's green and what's red across your projects, and which builds are running right now.
Search a build log, pull out its errors and warnings, or get a download link. Answers are sized to the question, not thousands of lines dumped into the chat.
The root cause, the last green build, the commits since, and whether the same failure appears elsewhere.
What changed between two builds, from the commits that went in to the results that came out.
Generate a SpeedrunAI summary of any build, to read or to share.
Start a build, rebuild a failure or cancel a stuck one, then follow the build until it finishes.
Connecting an AI app never gives it more than you have. Every request passes three checks: the connection allows it, your role allows it, and you approved it.
The AI works with your SpeedrunCI role. If you can't do something in the web app, your AI can't either.
Every change shows you a preview first: what will happen, where, and what it costs. Nothing runs until you approve it.
When an action needs extra care, you confirm it in SpeedrunCI by typing its name. The AI can't approve those by itself.
Credentials, signing keys, SSO setup and billing stay in the web app. When you need them, your AI links you to the right settings page.
Every call your AI makes is logged, and changes appear in your organization's audit log under the app's name.
Every app you've connected is listed under Connected apps. Narrow what it can reach, or revoke it whenever you like.
Add SpeedrunCI to your AI app, sign in the way you always do, and choose which organizations and toolsets it can use. Available on every plan, including trials.
https://mcp.speedrun.ci/mcpPaste it into your AI app as a custom connector or remote MCP server.
claude mcp add --transport http speedrun https://mcp.speedrun.ci/mcpThen type /mcp in Claude Code to sign in.
codex mcp add speedrun --url https://mcp.speedrun.ci/mcpThen run codex mcp login speedrun to sign in.
Claude (web, Desktop and mobile), ChatGPT, Claude Code, Codex, Cursor, VS Code, and any other app that supports remote MCP servers with OAuth sign-in.
No. It's available on every SpeedrunCI plan, including trials. Your plan's usual limits still apply: builds started from an AI app use build hours, and AI build summaries use SpeedrunAI credits, just as they do in the web app.
No. SpeedrunCI never writes to your repository. A coding agent such as Claude Code or Codex edits your local checkout, the way it already does, and you stay in control of what gets pushed.
No. It acts as you, with your SpeedrunCI role, in the organizations and toolsets you chose when you connected. Anything that changes something asks for your approval first.
They never pass through the chat. Credentials, signing keys, SSO setup and billing changes are web-only, and your AI gives you a link to the right page instead. Before log lines reach your AI, SpeedrunCI also removes text that looks like a secret.
A record of each call: the tool, the organization and project, whether it worked, and its arguments, with long text and anything named like a secret removed. It's kept for 180 days. SpeedrunCI doesn't store your conversations: your AI app sends only the tool name and its arguments.
Open Account settings, then Connected apps, and revoke it. You can also remove organizations or toolsets from an app without disconnecting it.