cs-agent-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| cwd | No | Working directory for child agents when no workspace root is provided |
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": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| cs_agent_capabilitiesA | Call first when considering delegation or heterogeneous execution. Lists configured local agent names, optional live availability probes, facade tools, and execution limits so you can decide whether and how to delegate. |
| cs_agent_createA | Create a managed child agent for a self-contained subtask that benefits from parallel work, a different agent runtime, specialization, or independent review. Call cs_agent_capabilities before choosing an agent; do not delegate trivial or tightly coupled work. |
| cs_agent_listA | Inspect managed agents in the caller's delegation subtree before creating duplicates or when coordinating parallel work. Supports filtering and cursor pagination; it cannot reveal sibling or ancestor subtrees. |
| cs_agent_statusA | Inspect one managed agent's lifecycle, queue, pending permission, last error, and runtime state. Use it to diagnose a delegated task before deciding to wait, respond to permission, cancel, or retry. |
| cs_agent_eventsA | Read or briefly wait for structured delegation events after an opaque cursor. Use for progress monitoring across agents or turns when a reply-oriented cs_agent_wait_message call is not sufficient. |
| cs_agent_sendA | Assign a self-contained task to a managed descendant agent. Include the objective, scope, constraints, expected deliverable, and verification criteria; the idempotency key makes retries safe. For parallel work, send all independent turns before waiting with cs_agent_wait_many. |
| cs_agent_get_messageA | Read one immutable delegated request or reply message when you already have its id. Use cs_agent_wait_message instead when waiting for an unfinished task. |
| cs_agent_wait_messageA | Preferred blocking wait after cs_agent_send for one turn. For multiple turns, send all independent turns first and use cs_agent_wait_many. Returns a terminal reply, a permission request requiring cs_agent_respond_permission, a terminal turn without a reply, or a bounded timeout. |
| cs_agent_wait_manyA | Wait for multiple turns after you send all independent turns first. Mode any returns all currently ready items; mode all waits until every turn is terminal but returns early for permissions or timeout. Accumulate ready items by turnId and continue with pendingTurnIds after an interrupted all wait. |
| cs_agent_get_turnA | Read one immutable-point-in-time turn snapshot by id. Use it for detailed state, revision, error, and permission diagnostics without waiting. |
| cs_agent_wait_turnA | Wait for a turn revision, permission request, or terminal state. Use this instead of cs_agent_wait_message when state transitions matter more than reply content. |
| cs_agent_respond_permissionA | Resolve a pending permission request surfaced by wait or status. Apply least privilege: prefer one-time approval unless repeated access is explicitly intended; rejection or cancel can stop delegated work. |
| cs_agent_cancelA | Cancel obsolete, unsafe, or no-longer-needed queued or active work, including unfinished descendant turns. This is a destructive control action; inspect status first when the outcome is uncertain. |
| cs_agent_destroyA | Destroy a managed agent after its work is complete or abandoned, optionally cascading through descendants and discarding its session. This releases the managed lifecycle and is destructive. |
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 14 tools
Tools have clearly distinct roles with detailed descriptions, but the multiple wait/read variants (wait_message, wait_turn, get_message, get_turn, events) create some overlapping surface that could cause misselection without careful reading.
All tools share the cs_agent_ prefix and use consistent lowercase_with_underscores, but the verb pattern is mixed: some are bare verbs (create, list, send, cancel, destroy) while others are verb_noun (get_message, wait_turn). This is still predictable and readable.
14 tools is well within the optimal range and each tool serves a distinct need in the delegation lifecycle: capabilities, creation, messaging, waiting, permissions, cancellation, and destruction. The count feels appropriate for the domain.
The lifecycle is well covered: create, list, status, send, wait, cancel, destroy, plus capability discovery and permission handling. Minor gaps exist such as no explicit update_agent or non-turn messaging tool, but these are likely out of scope and core workflows are fully supported.