Declare MCP servers once for every harness

You need a hall and an MCP server URL. This example declares Figma.

Add the server to ivar.json:

{
  "mcp": [
    { "name": "figma", "type": "http", "url": "https://mcp.figma.com/mcp" }
  ]
}

Then write the harness configs and authenticate:

ivar sync
ivar mcp auth figma --all-providers
ivar mcp status

What happens

• ivar sync rewrites each provider's native file from the declaration: .mcp.json for Claude Code, opencode.json for OpenCode and .omp/mcp.json for OMP.
• ivar mcp auth resolves the server from ivar.json, stores the OAuth client secret in .ivar/secrets/mcp.env and hands off to each provider's login path. With --all-providers it attempts every provider the hall lists and reports which failed.
• ivar mcp status reports credential status for each declared server and provider.
• Every session you start afterwards reads the same servers, whichever harness runs it.

A second server is one more entry in the mcp array and another ivar sync. A token never goes into ivar.json: the client secret lives in .ivar/secrets/mcp.env, which stays out of git, so the declaration is safe to commit and share with the team.

Run ivar mcp auth once per server, not once per harness. When a harness registers its own clients and the server rejects dynamic registration, ivar pre-registers a client for that provider first and reports the registration separately from the authentication.

Read the full walkthrough

Configure Figma MCP for OpenCode in one step shows the block and the auth step. The MCP guide covers transports and OAuth, and each harness has its own page: Claude Code, OpenCode and OMP.