The agent environment manager
ade captures the MCP servers, skills, instructions, settings, and plugins of 10 coding agents. Share the whole setup as one ADE.md file. Import any shared setup 1:1, see the plan, and apply it with a backup.
curl -fsSL https://raw.githubusercontent.com/woosal1337/ade/main/scripts/install.sh | sh$ ade apply acme/agentsSource acme/agents @ 9c54b4369133 (not trusted)acme/agents @ 9c54b4369133 does these things on this machine:! mcp github runs: npx -y @modelcontextprotocol/server-githubmcp docs connects to https://docs.example.com/mcpskill instructions for agents: review, write-commitTrust acme/agents @ 9c54b4369133? Type yes to continue: yesshared+ instructions AGENTS.md file, 317 bytes+ skill write-commit copy of ~/.cache/ade/trees/…claude+ mcp docs {"type":"http","url":"https://docs…"}pi+ mcp docs {"url":"https://docs.example.com/mcp"}Plan: 7 to create, 0 to update, 0 to delete. 0 already match.Applied 7 changes with 7 writes. Record: 20261012-090000-000-a1b2.
~/.claude.json{"numStartups": 41, yours"projects": { … }, yours"mcpServers": {"docs": { "type": "http", … } ade"local": { … } yours}}
One profile folder holds the setup. ade owns entries, not files: one server, one key, one skill. Every other key stays where it is.
claudesynced ~/.claude/settings.jsonsynced ~/.claude/hooks/extra ~/.claude/statusline.shprivate ~/.claude/.credentials.jsonruntime ~/.claude/sessions/42 synced, 5 extra, 0 unknown, 11 private.
Read the agent setup of this machine into one profile folder.
ade capture---ade: 1files: 74digest: "sha256:23aadac4…"---$ ade verify ADE.mdThe ADE.md is valid: 74 files.
Put the whole setup in one ADE.md, in any repository.
ade share --repo you/agents$ ade import acme/agents! mcp github runs: npx -y …skill instructions for agents: reviewTrust acme/agents @ 9c54b4369133? yesImported the profile into ~/.config/ade.Next: run `ade plan` to see the changes, …
Copy any shared setup into the profile folder on this machine.
ade import acme/agentscodex+ mcp github {"command":"npx",…}~ setting model "gpt-5.5" -> "gpt-5.6"! setting theme changed outside adePlan: 1 to create, 1 to update, 0 to delete.
See every change before ade writes a file. A plan never writes.
ade plan$ ade history20261012-091500-120-c3d4 apply 12 writes20261012-090000-000-a1b2 apply 7 writes$ ade revertReverted the record 20261012-091500-120-c3d4.
Make the machine match the profile. Each apply keeps a backup.
ade apply| Harness | Instructions | MCP | Skills | Settings | Plugins |
|---|---|---|---|---|---|
| Claude Code | Yes | Yes | Yes | Yes | Yes |
| Codex | Yes | Yes | Yes | Yes | Yes |
| Cursor | No | Yes | Yes | Yes | Yes |
| Grok | Yes | Yes | Yes | Yes | Yes |
| Pi | Yes | Yes | Yes | Yes | Yes |
| Gemini CLI | Yes | Yes | Yes | Yes | Yes |
| OpenCode | Yes | Yes | Yes | Yes | Partly |
| Zed | No | Yes | No | Yes | Yes |
| Claude Desktop | No | Partly | No | Yes | Partly |
| Antigravity | Yes | Yes | Yes | Yes | No |
● Synced ◐ Partly — Not available. ade never syncs sign-ins, tokens, sessions, history, or caches. Read the full support list.
The model is one more key. A profile can set it for each harness. ade does not change model names and does not sync sign-ins.
| Harness | File | Key | Example |
|---|---|---|---|
| Claude Code | settings.json | model | "opus" |
| Codex | config.toml | model | "gpt-6.1-sol" |
| Cursor | cli-config.json | model.modelId | "grok-4.7" |
| Grok | config.toml | models.default | "grok-4.5" |
| Pi | settings.json | defaultModel | "claude-opus-5-5" |
| Gemini CLI | settings.json | model.name | "gemini-3.1-pro-preview" |
| OpenCode | opencode.json | model | "anthropic/claude-opus-5-5" |
| Zed | settings.json | agent.default_model | { "provider": … } |
| Antigravity | settings.json | model | "Gemini 3.8 Flash (High)" |
GitHub shows it as a normal page: the install steps, what the setup runs and writes, the secret names, and every file. Each file has a SHA-256 hash, so a hand edit fails the check.
1Write one file from your profile
ade share --repo you/agents2Check it, then commit it
ade verify ADE.md3Others import it 1:1
ade import you/agents4Then they plan and apply it
ade apply1---2ade: 13name: "team"4harnesses: ["claude", "codex", "pi"]5secrets: ["GITHUB_TOKEN"]6files: 57digest: "sha256:4f1c9e…"8---910# team1112## Use this setup13ade apply acme/agents1415## What this setup does16| | Area | Item |17|---|---|---|18| ! | mcp | github runs: npx -y @modelcontextprotocol/server-github |19| | skill | instructions for agents: review |2021## Files2223```markdown ade-file path="skills/review/SKILL.md" sha256="9b1e…"24---25name: review26description: Review a change before a merge.27---28```
! mcp github runs: npx -y …! file writes ~/.npmrcskill instructions: reviewTrust acme/agents? yes
A Git source or a shared ADE.md needs your trust before ade imports or applies it. ade lists what it runs, writes, and reads. A change in any entry needs trust again.
"Authorization":"Bearer ${secret:GITHUB_TOKEN}""GITHUB_TOKEN":{ "from": "env" }
A profile names secrets and never holds their values. ade reads them at apply time from the environment, a file, a command, or the Keychain.
$ ade status! setting model changed outside ade$ ade revertReverted the record 2026…a1b2.
Every apply keeps backups and a record. A failed apply rolls back. ade revert restores the entries of one apply and leaves the rest.
curl -fsSL https://raw.githubusercontent.com/woosal1337/ade/main/scripts/install.sh | shNeeds Node.js 22 or later and Git. Add -s -- <owner>/<repo> to plan and apply a setup in the same step.
ade runs on macOS, Linux, and Windows. On Windows, ade copies files instead of making links. To set up another machine over SSH, run ade remote <host> apply. The other machine needs only Node.js.
1Capture this machine
ade capture2Write ADE.md, then push it to you/agents
ade share --repo you/agents3Import it on another machine
ade import you/agents4Show the plan
ade plan5Apply it
ade apply