Sub Agents MCP runs on your own machine — the client starts it, so there is no endpoint to ping. 217 installs a week from npm. Last commit 10 Sep 2026.
MCP server for delegating tasks to specialized AI sub-agents across coding tools
We read the source, 19 h ago · rules 3dff92dd89df
What this server is able to do. For an MCP server this is often the job itself — a terminal server runs commands because that is what it is for. Listed so you know what you are plugging in, not as an accusation.
childProcess = spawn(command, args, {
Is this your server and something here is wrong? Tell us — corrections are free and do not require a plan.
We found places where it runs commands, builds paths or queries from values it is given. None of that is a flaw by itself — it becomes one when the code changes, and code changes quietly between releases. We re-read it on every one.
This server runs on your own machine — install it with the package manager and the client starts it for you. Package name taken from the official registry entry.
claude mcp add sub-agents-mcp -- npx -y sub-agents-mcp
{
"mcpServers": {
"sub-agents-mcp": {
"args": [
"-y",
"sub-agents-mcp"
],
"command": "npx"
}
}
}
[mcp_servers.sub-agents-mcp]
command = "npx"
args = ["-y", "sub-agents-mcp"]
{
"mcpServers": {
"sub-agents-mcp": {
"args": [
"-y",
"sub-agents-mcp"
],
"command": "npx"
}
}
}
{
"mcpServers": {
"sub-agents-mcp": {
"args": [
"-y",
"sub-agents-mcp"
],
"command": "npx"
}
}
}
This one needs environment variables set before it will start:
AGENTS_DIR (Absolute path to the directory containing agent definition files (.md or .txt)), AGENT_TYPE (Type of AI CLI to use: 'cursor', 'claude', 'gemini', 'codex', 'command-code', 'glm', 'kimi', 'grok', 'antigravity', or 'opencode'), AGENT_PERMISSION (Approval/sandbox level for sub-agents: 'read-only', 'safe-edit' (default), or 'yolo'), AGENT_MODEL (Optional model override applied to every sub-agent execution by this MCP server), AGENT_EFFORT (Optional backend-specific reasoning effort or model variant. Supported by codex, claude, command-code, glm, kimi, grok, antigravity, and opencode.), CURSOR_API_KEY (API key for cursor-agent authentication (used only when AGENT_TYPE=cursor; passed via env, never via CLI args)), CLI_API_KEY (Provider API key for AGENT_TYPE=glm or kimi (passed to the claude binary via env, never via CLI args). Restart/reconnect the MCP server after changing it.), EXECUTION_TIMEOUT_MS (Agent execution timeout in milliseconds, as a positive whole number (default: 300000). The server refuses to start on an invalid value such as "5min".), SESSION_ENABLED (Enable session management to maintain execution history across agent calls (default: false)), SESSION_DIR (Directory path for storing session files (default: .mcp-sessions in current working directory)), SESSION_RETENTION_DAYS (Number of days to retain session history before automatic cleanup (default: 1)), AGENTS_SETTINGS_PATH (Path to CLI settings file/directory. Claude: --settings arg, Cursor: CURSOR_CONFIG_DIR, Codex: CODEX_HOME. GLM/Kimi/Gemini/Grok/Antigravity/OpenCode/Command Code not supported.).
The author declared them in the registry entry; get the values from the project itself.
MCP server mode for Qoder CLI — delegate coding tasks to local Qoder agents from any MCP client.
A growing Citadel of stack-specific docs, served as MCP tools for AI coding agents.
BridgeGuard MCP Server - Cross-chain bridge security audit tools for AI coding agents. Scan bri...
MCP server for Safari DevTools — browser debugging and automation for AI coding agents
App Store Connect MCP Server for AI coding agents
DeFiGuard MCP Server - DeFi protocol risk analysis tools for AI coding agents. Analyze token co...
Task management for AI coding agents — CQRS 2-gateway pattern, 200 ops across 10 domains
MCP server connecting AI agents to non-custodial staking data across 130+ networks.
Answers built from our own checks of this server.