MCP Agent Orchestrator
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
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 |
|---|---|
| get_task_statusB | Get the current status of a running task/job including progress, current step, and output. |
| list_jobsB | List all orchestrator jobs with their current statuses. |
| approve_checkpointB | Approve or reject a pending checkpoint in a pipeline execution. |
| get_configA | Get the current orchestrator configuration (sanitized, no API keys). |
| start_swarmA | Start an autonomous agent swarm to accomplish a complex goal. Returns a job ID immediately - the swarm runs in the background. Use get_task_status to monitor progress. |
| run_pipelineB | Execute a predefined pipeline by name. Returns a job ID immediately. |
| spawn_agentC | Spawn a new interactive agent with a specific role and system prompt. |
| send_agent_messageB | Send a message to a spawned agent and get its response. The agent will use its tools autonomously to complete the task. |
| list_agentsB | List all currently spawned interactive agents. |
| terminate_agentC | Terminate a spawned interactive agent. |
| invoke_sessionB | Spawn an independent specialist agent in an isolated session with its own context window, model/cost tier, system prompt, and tools. Supports preset personas (agent_preset) and model tiering (fast/balanced/heavy) to save cost while multiplying effective context. |
| invoke_sessions_parallelB | Fan out multiple tasks across separate isolated AI sessions in parallel, each with its own fresh context window and custom model/tier ("fast", "balanced", "heavy"). Multiplies effective context window capacity while keeping the main session clean. |
| send_session_messageB | Send an instruction or follow-up message to a specific agent session in the mesh. Can run asynchronously into the inbox or await the result directly. |
| check_inboxB | Check inbox for incoming messages, updates, or deliverables from spawned specialist sessions. |
| wait_for_inboxA | Event-driven wait for background child sessions to deliver a message to the inbox (no polling needed). Returns immediately if unread messages are already present. |
| list_sessionsB | List all agent sessions currently registered in the orchestrator mesh, including their roles, models, cost tiers, real-time statuses, and Context Firewall token savings. |
| get_session_transcriptB | Retrieve complete conversation logs, tool invocations, and inbox events for a specific session. |
| pause_sessionC | Pause execution of an active session in the mesh. |
| resume_sessionB | Resume execution of a paused session in the mesh. |
| terminate_sessionB | Terminate an active session and close its bilateral communication channel. |
| broadcast_session_messageB | Broadcast a directive or prompt simultaneously to all active child sessions (or all mesh sessions). |
| kill_all_sessionsB | Emergency stop and terminate all active sessions (or all children of a parent session). |
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 22 tools
The tool set splits orchestration across overlapping abstractions—agent vs. session, job vs. task vs. swarm—creating near-duplicate operations such as spawn_agent/invoke_session, send_agent_message/send_session_message, terminate_agent/terminate_session, and start_swarm/run_pipeline/invoke_sessions_parallel. Descriptions add context, but an agent can easily misselect which primitive is appropriate.
All tool names use snake_case verb_noun phrasing, e.g. get_task_status, list_jobs, spawn_agent, terminate_session, check_inbox. Adjective variants like kill_all_sessions and invoke_sessions_parallel are minor and still follow the same convention.
At 22 tools, the surface is heavy for an orchestration server, and much of that weight comes from duplicated agent/session concepts rather than clearly distinct capabilities. The count is defensible for advanced orchestration but feels bloated rather than tightly scoped.
Core lifecycle operations are covered: starting jobs, swarms, and pipelines; monitoring status; approving checkpoints; messaging agents and sessions; checking inboxes; retrieving transcripts; pausing/resuming; and terminating work. Minor gaps remain, such as listing available pipelines or canceling a specific non-session job, but most workflows have a viable path.