Skip to content
agentsync agentsync

Onboard a teammate

Goal: a teammate clones your repo and their AI agents pick up the project’s MCP servers and conventions with minimal setup.

The lever here is the project source tree: a .agentsync/ directory committed at the repo root, with the same layout as your user ~/.agentsync/.

  1. Scaffold the project tree at the repo root and commit it:

    Terminal window
    cd myrepo
    agentsync init --scope project    # creates ./.agentsync/

  2. Declare the project’s agents — required. Project scope renders only to the agents declared here, never each teammate’s user-scope agents, so everyone gets the same render from the committed tree:

    Terminal window
    agentsync agent add claude --scope project
    agentsync agent add opencode --scope project
    myrepo/.agentsync/agentsync.toml
    [agents]
    claude   = { enabled = true }
    opencode = { enabled = true }

  3. Add the project’s MCP server as its own file, and reference team secrets rather than embedding them so each teammate supplies their own value locally:

    myrepo/.agentsync/mcp/company-api.toml
    [server]
    type    = "stdio"
    command = "npx"
    args    = ["-y", "@company/mcp"]
    

    [server.env]
    COMPANY_TOKEN = "${secret:company.token}"

  4. Disable a noisy user plugin for this repo, if needed:

    myrepo/.agentsync/plugins/screenshot.toml
    [plugin]
    disabled = true

  5. Commit the whole .agentsync/ tree and document the one-time setup in your repo’s README: install agentsync, set the team secret, then apply from inside the repo.

What the teammate runs

Section titled “What the teammate runs”

  1. Install agentsync (install guide) and set the shared secret locally:

    Terminal window
    agentsync secrets set company.token --stdin < token.txt

  2. Apply from inside the repo — the overlay merges onto their user config:

    Terminal window
    cd myrepo
    agentsync apply --dry-run
    agentsync apply