Command line

Client compatibility

The CLI supports two setup paths: a local bridge for MCP clients that need to launch a command, and direct tool calls for scripts and scheduled jobs. If your client can connect to a remote MCP URL with an authorization header, use the direct configuration examples instead.

How to read the status

Verified

Confirmed with the CLI path shown in the row.

Expected

Expected to work from the client's documented MCP shape, but still worth verifying in your local setup.

Bridge clients

Use agentcentral mcp serve when a client needs a local command entry. The command reads your key from the environment or selected profile, then lets the client use agentcentral tools.

Local MCP clients

Verified

agentcentral mcp serve

Use the bridge when a client expects a local command instead of a remote MCP URL. The CLI reads your key from the environment or selected profile and forwards tool requests to agentcentral.

Claude Desktop

Expected

local command config

Use the generated config when you want Claude Desktop to launch the CLI locally. If you use Claude custom connectors instead, copy the signed connector URL from the dashboard and skip the CLI bridge.

Cursor

Expected

local command config

Use the generated config if your Cursor setup expects a local MCP command. If your Cursor build supports remote URLs with headers, direct setup is simpler.

OpenClaw

Expected

local command config

OpenClaw can use the bridge when you want a local command entry. For direct remote setup, use the OpenClaw quickstart.

Hermes Agent

Expected

local command config

Hermes can use the bridge through a local command entry. If you already have remote MCP config with headers, use that instead.

agentcentral mcp config claude   # print a starter config block
agentcentral mcp serve          # run the local bridge
agentcentral mcp test           # confirm the bridge can reach agentcentral

For setup details and local filter behavior, see the local MCP bridge.

Scripts and scheduled jobs

Use agentcentral tools call when no MCP client is involved. This is the path for terminal commands, cron jobs, and CI workflows.

Shell scripts

Verified

agentcentral tools call

Call any tool by name with inline JSON, stdin, or an input file. Use JSON output when another command needs to read the result.

cron and CI jobs

Verified

AGENT_CENTRAL_API_KEY + tools call

Pass the key through the job environment, write output to a file, and branch on the CLI exit code.

macOS, Linux, and Windows

Verified

Node.js runtime

The CLI runs anywhere Node.js 20+ is available. Local keychain storage is optional; environment-variable auth works everywhere.

For output formats, exit codes, and scheduled writes, see automation.

If your client supports headers

You do not need the CLI bridge. Connect directly to https://mcp.agentcentral.to/mcp with an Authorization header, or use a signed connector URL from the dashboard when the client requires one. See configuration examples.