MCP servers

The Model Context Protocollets an agent like Claude Code or Codex connect to external tools and data sources — filesystems, databases, issue trackers, your own services — through a standard interface. CodeGrid doesn't replace that protocol; it gives you one place to manage the servers each agent can reach.

Open the MCP Manager from the command palette (⌘K→ “MCP”) or Settings. It lists the MCP servers configured for your agents and lets you:

• Add a server (command + args, or paste a JSON config block).
• Toggle servers on/off per agent.
• Remove servers you no longer need.

Edits are written back to each agent's own MCP config, so the agents pick them up the same way they would from the CLI.

Most servers are a command CodeGrid launches plus its arguments. You can fill those fields directly, or paste a Claude-Code-style JSON block — the manager merges it into the agent's config. For example, to add the filesystem server scoped to a project directory:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/me/projects"]
    }
  }
}

Each entry under mcpServers is keyed by a name you choose (here filesystem), with a command and an args array. Servers that need secrets can also take an env object.

CodeGrid's own Agent Bus is delivered as an MCP server (codegrid-agent-bus). Enabling collaboration in onboarding — or running its setup— registers it into each agent's MCP config for you.