Everruns plugins

Plugins that teach your coding agent to use Everruns, the platform for durable AI agents. Install one, sign in once, and ask your agent to build, run, debug or ship Everruns agents for you.

> Create an Everruns agent that summarizes Hacker News every weekday morning,
  run it once, and post it to our Slack.

Works in Claude Code, Codex, Cursor and Gemini CLI, and in any host that reads the portable Agent Plugins format.

What you get

The everruns plugin connects your agent to the Everruns MCP server (https://app.everruns.com/mcp, OAuth sign-in in the browser) and adds skills that load only when a task needs them:

Skill	Your agent can…
everruns	Use Everruns at all: concepts, the MCP tools and the everruns <noun> <verb> command language, safe changes, secrets, links, and which skill to load next
everruns-run-agent	Hand a task to one of your Everruns agents, follow it, pass files in and out, and bring the result back
everruns-build-agent	Create or change an agent from a plain-language description, give it knowledge and skills, and test it with a real message
everruns-debug-session	Explain why a session failed, stalled or gave a bad answer, fix it, and keep it fixed
everruns-evaluate-agent	Prove an agent works with evals built from real sessions, compare models, and score live traffic
everruns-ship-agent	Put an agent in front of people: Slack, web chat, A2A, webhooks, schedules and GitHub events, with a budget and a kill switch
everruns-operate	Answer what was spent, what is broken and who changed what across your organization
everruns-agents-as-files	Keep agents as agent.toml folders in your repository, and validate, diff, import and export them
everruns-sdk	Call Everruns from your app with the Python, TypeScript and Rust SDKs
everruns-framework	Run agents inside your own Rust program with the everruns crate or everruns-serve

Install

You need an Everruns account. Sign up at https://app.everruns.com; the first tool call opens the sign-in page in your browser.

Claude Code

/plugin install everruns --marketplace everruns/plugins

On Claude Code older than 2.1.275, add the marketplace first: /plugin marketplace add everruns/plugins, then /plugin install everruns@everruns.

Codex

codex plugin marketplace add everruns/plugins
codex plugin add everruns@everruns

Or run /plugins in Codex and pick Everruns. Start a new session after installing.

Cursor

Open Settings > Plugins, paste https://github.com/everruns/plugins, and install Everruns. From the Cursor CLI: cursor-agent plugin marketplace add everruns/plugins, then /plugin and pick it from the Marketplace tab.

Gemini CLI

gemini extensions install https://github.com/everruns/plugins

Any other agent

• Agent Plugins hosts (GitHub Copilot, VS Code, Kiro and others): point the host at plugins/everruns/. Its plugin.json, mcp.json and skills/ follow Agent Plugins 1.0.
• MCP only: add https://app.everruns.com/mcp as a remote (Streamable HTTP) server named everruns.
• Skills only: copy plugins/everruns/skills/* into your agent's skills folder (for example .claude/skills/ or .agents/skills/).

First run

1. Ask: "Who am I on Everruns?" Your host opens the browser to sign in.
2. Then ask for real work: "Ask my research agent to summarize this repo", "Why did https://app.everruns.com/sessions/… fail?", "How much did our agents spend this week?", "Export the triage agent into agents/triage".

Works even better with the CLI

The plugin needs nothing but the MCP server: it works in chat apps without a shell. Under the hood, everything Everruns can do is one command language, everruns <noun> <verb> --flags, which the MCP server runs for your agent. If the everruns CLI is installed and signed in, the skills use it directly instead, which is faster, needs no tool schemas, and adds local file sync.

brew tap everruns/tap && brew install everruns
everruns login

Self-hosted Everruns

The plugin points at Everruns Cloud. For your own deployment:

• CLI: everruns --api-url https://<your-host>/api login, or set EVERRUNS_API_URL.
• MCP: register your server under the same name, for example claude mcp add --transport http everruns https://<your-host>/mcp, and keep the skills.

Security and privacy

• Sign-in uses OAuth 2.1 with your Everruns account. The plugin contains no keys and stores nothing.
• Your agent acts as you, with your permissions in each organization.
• Secrets never pass through the model: the skills use links where you type the value into Everruns directly.
• Every change your agent makes is recorded in Everruns history with the reason it gave.

Report a vulnerability to security@everruns.com, not in a public issue.

Repository layout

plugins/everruns/
  plugin.json, mcp.json         portable Agent Plugins manifest and MCP config (hand-written)
  skills/<name>/SKILL.md        the skills (hand-written)
  .claude-plugin/ .codex-plugin/ .cursor-plugin/ .mcp.json   host manifests (generated)
  assets/                       logo and icon
  evals/                        scenario evals against a mocked Everruns organization
catalog.json                    marketplace listing and per-host metadata (hand-written)
.claude-plugin/ .agents/plugins/ .cursor-plugin/   marketplaces (generated)
gemini-extension.json, skills   Gemini CLI entry point (generated)
scripts/build.mjs               generator and checks
scripts/check-*.mjs             skills match the live MCP tools and commands
scripts/eval.mjs                runs the scenario evals with and without the skills

Contributing

Edit only the hand-written files, then run:

node scripts/build.mjs           # regenerate host files
node scripts/build.mjs --check   # what CI runs
node scripts/check-commands.mjs  # every `everruns …` command in a skill exists

Scenario evals

plugins/everruns/evals/ holds tasks a user would give a coding agent: hand work to an agent, explain a failed session, report spend, set up regression evals, handle a pasted API key, publish publicly, and write an SDK script. Each runs against a mocked Everruns MCP server, so no account is touched. The script runs every case three ways: with the plugin, with the MCP server but no skills, and with no plugin, so the table shows what the skills themselves add.

ANTHROPIC_API_KEY=... node scripts/eval.mjs --runs 3 --model claude-sonnet-5-5

It needs Claude Code (claude plugin eval) and spends real model tokens, so it is run by hand before a release rather than in CI. When you change a skill, run the cases it affects (--case <name>).

Skills follow the Agent Skills format: a short description that says when to load the skill, a body under 500 lines, and no API detail the server already documents (--help and discover are the source of truth for flags). Bump version in plugin.json when a skill or manifest changes so hosts pick up the update.

Issues and pull requests are welcome. For Everruns itself, see everruns/everruns.

License

MIT