mcp-agent mcp

v1.0.0 · MCP Tool · ai · registry.pascalai.org

Spawn and coordinate AI sub-agents via MakerAI AgentScheduler (_PAI_Spawn). Define an agent's goal, available tools, memory namespace and LLM model, then run it synchronously (await result) or asynchronously (fire-and-forget with instance ID).

agentspawnorchestrationmakerAIautonomousgoal

MakerAI Pipeline

Input Parameters

ParameterTypeDescription
operationrequired string 'spawn': create and start an agent; 'status': get agent run status; 'cancel': stop a running agent; 'send': send a message to running agent via blackboard; 'await': wait for agent completion and get result. One of: spawn, status, cancel, send, await.
agent_nameoptional string Agent definition name (spawn). Must match a registered agent type.
goaloptional string Natural language goal for the agent to accomplish (spawn).
toolsoptional array[string] MCP tool names available to the agent (spawn). E.g. ['mcp-fetch', 'mcp-sqlite'].
memory_namespaceoptional string Agent memory namespace. Default: auto-generated from instance ID.
modeloptional string LLM model for the agent. Empty = router default.
max_stepsoptional integer Max reasoning steps before forced stop. Default: 20. Default: 20.
asyncoptional boolean If true, returns immediately with instance_id. If false, waits for completion. Default: false. Default: False.
argsoptional object Initial arguments passed to the agent (spawn).
instance_idoptional string Agent instance ID (for status, cancel, send, await).
messageoptional object Message payload to send to running agent (send operation).
topicoptional string Blackboard topic for inter-agent messaging (send).

Output Fields

FieldTypeDescription
operation string
instance_id string Unique ID of the spawned agent instance.
status string
result any Agent output on completion.
steps_taken integer
tools_used array[string]
tokens_used integer
elapsed_ms integer
error string

Examples

Spawn a research agent synchronously

// Input
{
  "operation": "spawn",
  "agent_name": "researcher",
  "goal": "Find the top 5 open source vector databases by GitHub stars and summarize their key features",
  "tools": [
    "mcp-fetch",
    "mcp-brave-search",
    "mcp-summarize"
  ],
  "max_steps": 10
}

// Output
{
  "instance_id": "agent_7f3a2b",
  "status": "completed",
  "result": {
    "findings": "1. Chroma (22k stars)... 2. Qdrant (18k stars)..."
  },
  "steps_taken": 7,
  "tools_used": [
    "mcp-brave-search",
    "mcp-fetch"
  ],
  "tokens_used": 3420,
  "elapsed_ms": 8240
}

Install & Discovery

Install

ppm install mcp-agent

Get JSON Schema

GET /v1/packages/mcp-agent/1.0.0/schema

Discover by keyword

GET /v1/mcp/discover?q=agent
Discovery hint: Install with ppm install mcp-agent or invoke remotely via POST /v1/invoke/mcp-agent on the MCP Service.

PascalAI Usage

uses toolslib;
var Tool := LoadTool('mcp-agent');
var R := Tool.Call(JsonObj([
  'operation','spawn',
  'agent_name','researcher',
  'goal','Summarize latest AI news',
  'tools',JsonArr(['mcp-fetch','mcp-brave-search'])
]));
Writeln(R['result']);