CLI Subagents MCP Server
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| CLAUDE_CONFIG_DIR | No | Directory of the Claude Code config folder. Claude Code's settings.json is looked up at ~/.claude/settings.json, or the one in $CLAUDE_CONFIG_DIR. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| subagent_startA | Start a subagent: another installed agent CLI works on one task in the background, under one of your accounts, with your shared MCP servers, plugins, skills and settings. Returns a job_id at once (or the result, if it finishes within wait_seconds). Give it a complete, self-contained task: goal, scope, files, and how to check the result. csa enforces the account limits itself: if a start is refused, follow the suggestion in the message. |
| subagent_waitA | Wait until one of the given jobs finishes, or a subagent sends a message or question, then return what happened. In Claude Code a long wait moves to the background on its own and you get a notification when it returns, so call this after starting jobs instead of polling. |
| subagent_messageA | Send a message to a subagent. While it runs: a follow-up or an answer to its question (set reply_to to the question's id). After it finished: continues the same conversation as a new job and returns that job. |
| subagent_statusC | State and progress of a job. Use wait_seconds instead of polling in a tight loop. |
| subagent_resultC | Final answer, outcome, account, session and cost of a finished job. |
| subagent_cancelB | Stop a running job and every process it started. |
| subagent_listD | Recent jobs, newest first. |
| subagent_accountsC | Installed runtimes, their accounts, login state, last known usage and cooldowns. Makes no model calls. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 8 tools
Most tools have clearly distinct purposes: start, cancel, list, and accounts are unambiguous. However, wait, status, and result all relate to monitoring a job and could be confused—wait blocks for completion/messages, status is a snapshot, result is final output—but descriptions clarify intended usage.
All tools use the consistent subagent_ prefix with snake_case and clear action verbs. No deviations in style or convention.
8 tools is well-scoped for a subagent orchestration server, covering the full job lifecycle without redundancy. Each tool earns its place.
The surface covers starting, monitoring (wait/status/list), interacting (message), retrieving results, cancelling, and account management. A minor gap is the lack of a tool to retrieve full transcript/logs or delete jobs, but core workflows are covered.