pegboard install mcp linear=…Every tool in its place. On every agent.
Pegboard is Configuration as Code for coding agents. Declare your MCPs, skills, and context in one file, and it writes each agent's native config. You see the plan before anything changes.
curl -fsSL https://usepegboard.dev/install.sh | shRead the quick start
Runs on Linux and macOS, with binaries for Windows.
Every agent keeps its tools somewhere else. Claude Code uses JSON, Codex uses TOML, Hermes uses YAML. Pegboard translates one desired state into all of them with tested code, so neither you nor your agent has to do it by hand.
Ask your agent. It runs Pegboard.
Install the built-in skill once, and any of your agents can add, remove, and move tools without rewriting config files itself. It shows you the plan first, and asks you about conflicts. How the skill works
- "Add the Linear MCP to all my agents"
- "Install this skill everywhere"
pegboard install skill owner/repo@skill- "Remove Notion from my agents"
pegboard uninstall mcp notion
See what moved. Put it back.
On a workshop shadow board, every tool has a painted silhouette, so an empty spot stands out. pegboard.yaml is your silhouettes, and it works the same way.
- Plan what changes
pegboard plan- Shows every create, update, removal, and conflict. It writes nothing.
- Apply it
pegboard apply- Writes each agent's files, and a second run changes nothing.
- Catch drift
pegboard drift --exit-code- Exits with 2 when a managed file was edited, so CI can catch it.
Copy between agents
Already set up in one agent? Name it, whichever it is, then the agents that should get the same MCPs and skills. You don't need a pegboard.yaml, and the first agent doesn't change. Secrets arrive as references, and you choose whether its context goes before, after, or in place of what each agent has. See what it copies
Safe on your config
- It leaves your edits alone
- Pegboard only touches the keys, files, and links it created. Everything else in the file stays as you wrote it.
- It backs up before it writes
- Every file it changes is written atomically, with a copy saved in
~/.local/state/pegboard/backups. - Secrets stay out of the YAML
- Reference them with
${env:NAME}or${file:path}. If an agent can only store plain text, the plan warns you first. - Skills stay pinned
pegboard.lockrecords the commit and hash of each git skill. If the content changes,applyfails.
Nine agents today
All of them get MCPs and skills, each in the place that agent expects. See every path
| Agent | Global MCP config | Context |
|---|---|---|
| Claude Code | ~/.claude.json | CLAUDE.md bridge |
| Codex | ~/.codex/config.toml | AGENTS.md |
| opencode | ~/.config/opencode/opencode.json | AGENTS.md |
| Hermes | ~/.hermes/config.yaml | SOUL.md (its persona) |
| Pi | ~/.pi/agent/mcp.json | AGENTS.md |
| OMP (oh-my-pi) | ~/.omp/agent/mcp.json | .omp/AGENTS.md |
| OpenClaw | ~/.openclaw/openclaw.json | Workspace AGENTS.md |
| Cursor | ~/.cursor/mcp.json | Project AGENTS.md |
| Gemini CLI | ~/.gemini/settings.json | GEMINI.md |
Skip the YAML
When you need one tool right now, ad-hoc installs get the same plan, state, and backups. To keep them, pegboard adopt moves them into pegboard.yaml.
Share it with a team
Commit pegboard.yaml and pegboard.lock. Everyone runs pegboard apply after cloning, the same way they'd run npm install. Set up CI and policies
Read the docs
From the first install to every flag and file path. Open the docs
Questions
Every tool in its place. On every agent.
curl -fsSL https://usepegboard.dev/install.sh | sh