Your first sync

This is the greenfield path: start clean, end with an MCP server live in two agents. About five minutes.

1. Create ~/.agentsync/ and its layout.

   agentsync init

2. Register the agents you use.

   agentsync agent add claude
   agentsync agent add opencode

3. Add an MCP server once — it will fan out to both agents.

   agentsync mcp add github \
     --command npx \
     --args "-y,@modelcontextprotocol/server-github"

4. Preview before writing anything. Always safe; never touches disk.

   agentsync apply --dry-run

5. Apply for real.

   agentsync apply

Confirm it landed

The one server you added now lives in both agents’ native configs, each in its own shape:

jq '.mcpServers.github' ~/.claude.json
jq '.mcp.github'        ~/.config/opencode/opencode.json

`apply --dry-run` is your friend

It prints the full translation report — what lands natively (✓), what’s projected with loss (◐), and what’s skipped (✗) — without writing a byte. Run it before every real apply until you trust the output.

What just happened

flowchart LR
    SRC["~/.agentsync/mcp/github.toml<br/>(one file you own)"]
    SRC -->|apply| CLAUDE["~/.claude.json<br/>.mcpServers.github"]
    SRC -->|apply| OC["opencode.json<br/>.mcp.github"]

You wrote one small TOML file. apply rendered it into each agent’s native schema — Claude’s mcpServers, OpenCode’s mcp (with its different transport and command shape). That’s the fan-out.

Next steps

The daily loop The four commands you'll actually use day to day.

Add a plugin Install once from a marketplace; every agent gets the components it understands.

Reference a secret Stop hard-coding tokens. Pull them from an age-encrypted vault at apply time.

Already have configs? Import an existing Claude / OpenCode setup instead of retyping it.