RELATED 01
Claudish
MCP mode of the Claudish CLI: send prompts to non-Anthropic models and run them as Claude Code sessions
1.0kGitHub stars
1.9k/wknpm downloads
58popularity score
Crewly
Runs and steers a local team of Claude Code, Gemini CLI and Codex agents from your MCP client
By Steve's Prompt (individual) TypeScript Local · stdio AI and MLDeveloper tools
57 popularity score ★ 92 ↓ 3.7k/wk
Connect your client
Local (npm)
Add in one click
Opens the app, which asks before it adds the server.
Or add it by command or config
Claude CodeTerminal
claude mcp add crewly -- npx -y crewly mcp-serverClaude Desktopclaude_desktop_config.json
{
"mcpServers": {
"crewly": {
"command": "npx",
"args": [
"-y",
"crewly",
"mcp-server"
]
}
}
}Cursor~/.cursor/mcp.json
{
"mcpServers": {
"crewly": {
"command": "npx",
"args": [
"-y",
"crewly",
"mcp-server"
]
}
}
}VS Code.vscode/mcp.json
{
"servers": {
"crewly": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"crewly",
"mcp-server"
]
}
}
}Codex~/.codex/config.toml
[mcp_servers.crewly]
command = "npx"
args = ["-y", "crewly", "mcp-server"]Gemini CLITerminal
gemini mcp add --scope user crewly npx -- -y crewly mcp-serverWhat it does
Crewly is a local orchestration app for AI coding agents: it starts Claude Code, Gemini CLI, Codex or OpenCode as separate terminal sessions, gives each a role (developer, QA, product manager, orchestrator) on a project folder, and shows them in a web dashboard with task boards and shared memory. Its npm package also contains an MCP server, started with `crewly mcp-server`, which speaks stdio and lets another assistant manage those teams from outside.
The README does not mention the MCP server at all; it is documented only in the source. The comment in cli/src/commands/mcp-server.ts lists six tools, but the tool definitions at v1.20.347 register twelve: besides listing and creating teams, assigning tasks, reading status and searching team memory, there are five tools that manage Crewly's encrypted credential store (add an API key, import a Google login from the Gemini CLI Workspace extension, list, delete) and one that runs an installed Crewly skill with chosen credentials.
It suits a setup where one assistant acts as a coordinator and hands work to the agent crew Crewly already runs on the same machine.
What it needs
| Runtime | Node.js 22+ |
|---|
Node.js 22 or later with npm (the README says Node 20 lacks a prebuilt better-sqlite3 binary); python3, make and g++ on Alpine/musl, old glibc or other architectures. A normal user account, not root. Crewly set up with `crewly init` (needs jq) and at least one agent CLI installed and signed in: Claude Code, Gemini CLI (with GEMINI_API_KEY) or Codex (OpenAI key). `crewly start` must be running for task assignment.
Limits
crewly_send_message does not deliver anything at v1.20.347: the handler only returns a 'queued' confirmation with a conversation id (backend/src/services/mcp-server.ts). crewly_assign_task posts to the local Crewly backend and returns an error when it is not running. Tasks are only picked up by agents Crewly has already started. The docs' example command is `npx crewly mcp-server` without -y, which can stop at npx's install prompt; the snippet here adds -y. The MCP server is not covered in the README.
Who it suits
Developers who already run Crewly to coordinate several coding agents and want Claude Code, Cursor or another MCP client to create teams, hand out tasks and check progress without opening the dashboard. Without a Crewly install and signed-in agent CLIs it does nothing useful, so it is not a general agent or task tool. Because it can store and delete API keys and run skills with them, it fits a personal machine better than a shared one. It is a community project, not made by Anthropic, Google or OpenAI.
Tools 12
crewly_get_teams- List Crewly teams with their members, roles and agent status, optionally one team.
crewly_create_team- Create a team with a name, description and members, each with a role and agent runtime.
crewly_assign_task- Hand a task to one team member through the running Crewly backend.
crewly_get_status- Report the status of all teams, one team or one member.
crewly_recall_memory- Search the agents' and project's stored knowledge, scoped by agent or project path.
crewly_send_message- Accept a message for the orchestrator and return a conversation id.
crewly_credential_list- List stored API keys and Google logins by metadata, without secret values.
crewly_credential_add_api_key- Save an API key for a provider in the encrypted credential store.
crewly_credential_oauth_import_gemini_cli- Copy the Google login from the Gemini CLI Workspace extension into the store.
crewly_credential_clear_gemini_cli_file- Delete the Gemini CLI Workspace extension's cached token file.
crewly_credential_delete- Permanently remove a stored credential.
crewly_execute_skill- Run an installed Crewly skill, binding stored credentials to its slots, and return redacted output.