AI agents
buildtree ships an MCP server (@buildtree/mcp). Add it to your coding agent and you can say "set up buildtree for this app and give me a QR code for an Android build". The agent logs in, creates a project, writes the build config, runs the build, uploads the artifact and hands back install links plus the QR image. It can also read the feedback testers sent from install pages and start fixing it.
Requires Node 20 or later on the machine running the agent.
Add it to your agent
Claude Code
claude mcp add --transport stdio buildtree -- npx -y @buildtree/mcp
Add --scope user to make it available in every project. A project-level .mcp.json works too:
{ "mcpServers": { "buildtree": { "command": "npx", "args": ["-y", "@buildtree/mcp"] } } }
Codex CLI (~/.codex/config.toml)
[mcp_servers.buildtree]
command = "npx"
args = ["-y", "@buildtree/mcp"]
tool_timeout_sec = 1800
Codex defaults to a 60 second tool timeout. Raise it as above so buildtree_build can finish, or let the agent run the build in its own shell and only upload through the server.
Gemini CLI (~/.gemini/settings.json)
{ "mcpServers": { "buildtree": { "command": "npx", "args": ["-y", "@buildtree/mcp"], "timeout": 1800000 } } }
Claude Desktop: download the .mcpb file from the latest release and double-click it. Claude Desktop installs it with one click and asks for an optional API token; leave it empty to sign in through the browser.
Cursor, Windsurf: the same npx -y @buildtree/mcp command in .cursor/mcp.json or mcp_config.json. Claude Desktop also accepts it in claude_desktop_config.json if you prefer config files.
Logging in
The first time, ask the agent to log in. It calls buildtree_login, which opens a browser page where you approve the connection, then buildtree_login_status. The token is stored where the buildtree CLI stores its own, so one login serves both.
In CI, or on a machine without a browser, set BUILDTREE_TOKEN to an API token from the dashboard's API Tokens page. Tokens never appear in tool output.
What the agent can do
| Tool | What it does |
|---|---|
buildtree_whoami | Account, organization and plan. The agent calls this first. |
buildtree_login, buildtree_login_status | Browser login, shared with the CLI. |
buildtree_list_projects, buildtree_create_project | One project per app. |
buildtree_detect_project | Looks at the app directory (Expo, React Native, Flutter, native) and suggests build commands and artifact paths. Runs nothing. |
buildtree_write_config | Writes buildtree.config.json. |
buildtree_build | Runs the configured build with progress updates, records the artifact. |
buildtree_upload | Uploads an .apk, .aab or .ipa; returns pinned, folder and release URLs plus a QR code image. |
buildtree_list_builds, buildtree_get_install_links | Browse builds, get links and QR for any build. |
buildtree_list_feedback | Read what testers reported, screenshots included. |
The server also exposes buildtree://guide (the workflow) and buildtree://frameworks/{expo|react-native|flutter|android|ios} (build recipes) as resources, and a setup prompt.
Tell your agent about buildtree
If you keep an AGENTS.md or CLAUDE.md in the repo, a few lines save the agent a search:
## Distributing builds
Use the buildtree MCP server to upload QA builds and get install links.
If it is not installed: `claude mcp add --transport stdio buildtree -- npx -y @buildtree/mcp`.
Upload to the `dev` environment; pass the git branch as `branch` for feature work.
Source code
The server is open source (MIT) at github.com/engineerdeep/buildtree-mcp. It is also listed in the official MCP registry and on Smithery.
Security notes
- Feedback text comes from testers. The server labels it as untrusted input and the agent should treat it as data, not instructions.
- The server writes nothing to your project except
buildtree.config.json, and only when asked. - Build logs are saved under your system temp directory (
buildtree-mcp/).
Last updated