Skip to content

mcp.md

Latest commit

 

History

History
509 lines (413 loc) · 12.1 KB

mcp.md

File metadata and controls

509 lines (413 loc) · 12.1 KB

MCP Integration

Working examples for connecting MCP clients to OpenIntent and consuming external MCP tools from OpenIntent agents.

Connecting Claude Desktop

Add the OpenIntent MCP server to your Claude Desktop configuration:

{
  "mcpServers": {
    "openintent": {
      "command": "npx",
      "args": ["-y", "@openintentai/mcp-server"],
      "env": {
        "OPENINTENT_SERVER_URL": "http://localhost:8000",
        "OPENINTENT_API_KEY": "your-api-key",
        "OPENINTENT_MCP_ROLE": "operator"
      }
    }
  }
}

After restarting Claude Desktop, you can manage intents, events, and channels through natural language:

from openintent import OpenIntentClient

client = OpenIntentClient(base_url="http://localhost:8000", api_key="dev-key")
intent = client.create_intent(
    title="Q1 Analysis",
    description="Analyze Q1 sales performance",
    initial_state={"phase": "planning"},
)
print(f"Created: {intent.id}")

Claude sees the same intent through MCP and can update its state:

// Claude calls openintent_update_state
{
  "intent_id": "intent_01abc",
  "version": 1,
  "state_patch": {
    "phase": "in_progress",
    "assigned_to": "data-team"
  }
}

Read intent state as a resource:

openintent://intents/intent_01abc/state
→ { "phase": "in_progress", "assigned_to": "data-team" }

Browse the event log:

openintent://intents/intent_01abc/events
→ [{ "event_type": "state_patched", "actor": "mcp-agent", ... }]

Role-Based Access Examples

Read-Only Dashboard

A monitoring tool that can observe protocol state but never modify it:

{
  "mcpServers": {
    "openintent-monitor": {
      "command": "npx",
      "args": ["-y", "@openintentai/mcp-server"],
      "env": {
        "OPENINTENT_SERVER_URL": "https://openintent.example.com",
        "OPENINTENT_API_KEY": "monitor-api-key",
        "OPENINTENT_MCP_ROLE": "reader"
      }
    }
  }
}

Available tools: get_intent, list_intents, get_events, get_messages (4 tools).

Worker Agent

An agent that creates intents, updates state, logs events, and sends messages — but cannot manage lifecycle, leases, or agent assignments:

{
  "mcpServers": {
    "openintent-worker": {
      "command": "npx",
      "args": ["-y", "@openintentai/mcp-server"],
      "env": {
        "OPENINTENT_SERVER_URL": "https://openintent.example.com",
        "OPENINTENT_API_KEY": "worker-api-key",
        "OPENINTENT_MCP_ROLE": "operator"
      }
    }
  }
}

Available tools: all read + write tier tools (10 tools).

Trusted Orchestrator

A coordinator with full access to all protocol primitives:

{
  "mcpServers": {
    "openintent-orchestrator": {
      "command": "npx",
      "args": ["-y", "@openintentai/mcp-server"],
      "env": {
        "OPENINTENT_SERVER_URL": "https://openintent.example.com",
        "OPENINTENT_API_KEY": "orchestrator-api-key",
        "OPENINTENT_MCP_ROLE": "admin"
      }
    }
  }
}

Available tools: all 16 tools.

Restricted Operator

An operator further limited to specific tools using the allowlist:

{
  "server": {
    "url": "https://openintent.example.com",
    "api_key": "restricted-api-key",
    "agent_id": "data-writer"
  },
  "security": {
    "tls_required": true,
    "role": "operator",
    "allowed_tools": [
      "openintent_create_intent",
      "openintent_update_state",
      "openintent_get_intent",
      "openintent_list_intents"
    ],
    "audit_logging": true
  }
}

Available tools: only the 4 listed tools (even though operator grants 10).

MCPTool in @Agent (First-Class Integration)

LLM-Powered Agent with OpenIntent MCP Tools

An agent that uses MCPTool to automatically discover and call OpenIntent tools through MCP:

from openintent import Agent, MCPTool, on_assignment

@Agent("analyst", model="gpt-4o", tools=[
    MCPTool(
        server="npx",
        args=["-y", "@openintentai/mcp-server"],
        role="operator",
        env={"OPENINTENT_SERVER_URL": "http://localhost:8000"},
    ),
])
class Analyst:
    @on_assignment
    async def work(self, intent):
        return await self.think(intent.description)

Analyst.run()

At startup, the SDK connects to the MCP server, discovers all tools the operator role permits (10 tools), registers them as native ToolDef entries, and the LLM can call them during self.think().

Read-Only Watcher via mcp:// URI

from openintent import Agent, on_assignment

@Agent("watcher", model="gpt-4o", tools=[
    "mcp://npx/-y/@openintentai/mcp-server?role=reader",
])
class Watcher:
    @on_assignment
    async def work(self, intent):
        return await self.think("Summarize: " + intent.description)

Mixed Tools — Local + Multiple MCP Servers

from openintent import Agent, MCPTool, tool, on_assignment

@tool(description="Search the web.", parameters={
    "type": "object",
    "properties": {"query": {"type": "string"}},
    "required": ["query"],
})
async def web_search(query: str) -> dict:
    return {"results": [f"Result for {query}"]}

@Agent("researcher", model="gpt-4o", tools=[
    web_search,
    MCPTool(
        server="npx",
        args=["-y", "@openintentai/mcp-server"],
        role="operator",
        allowed_tools=["get_intent", "list_intents", "set_status"],
    ),
    MCPTool(
        server="npx",
        args=["-y", "@modelcontextprotocol/server-filesystem", "/data"],
        allowed_tools=["read_file", "list_directory"],
    ),
])
class Researcher:
    @on_assignment
    async def work(self, intent):
        return await self.think(intent.description)

Coordinator with MCP-Backed Governance

from openintent import Coordinator, MCPTool, on_assignment

@Coordinator(
    coordinator_id="ops-lead",
    agents=["analyst", "writer"],
    model="gpt-4o",
    tools=[
        MCPTool(
            server="npx",
            args=["-y", "@openintentai/mcp-server"],
            role="admin",
            env={"OPENINTENT_SERVER_URL": "http://localhost:8000"},
        ),
    ],
)
class OpsLead:
    @on_assignment
    async def work(self, intent):
        return await self.think(
            f"Coordinate: {intent.description}"
        )

Python Agent with MCP Tools (Low-Level Bridge)

An agent that reads data from a filesystem MCP server using the low-level MCPBridge:

#!/usr/bin/env python3
"""
Agent that uses a filesystem MCP server to read input data.

Prerequisites:
    pip install openintent mcp
    npm install -g @modelcontextprotocol/server-filesystem

Usage:
    python mcp_agent.py
"""

import asyncio
from openintent.mcp import MCPServerConfig, MCPSecurityConfig, MCPBridge

async def main():
    config = MCPServerConfig(
        name="filesystem",
        command="npx",
        args=["-y", "@modelcontextprotocol/server-filesystem", "/data"],
        allowed_tools=["read_file", "list_directory"],
        security=MCPSecurityConfig(audit=True),
    )

    bridge = MCPBridge()
    bridge.add_server(config)
    await bridge.connect_all()

    tools = await bridge.list_all_tools()
    print("Available tools:")
    for server, server_tools in tools.items():
        for tool in server_tools:
            print(f"  {server}/{tool['name']}")

    listing = await bridge.invoke(
        "filesystem", "list_directory", {"path": "/data"}
    )
    print(f"\nDirectory listing: {listing['content']}")

    result = await bridge.invoke(
        "filesystem", "read_file", {"path": "/data/input.txt"}
    )
    print(f"\nFile contents: {result['content']}")

    await bridge.disconnect_all()

asyncio.run(main())

Using MCPToolProvider as a context manager for a single server:

from openintent.mcp import MCPToolProvider, MCPServerConfig

async def read_with_context_manager():
    config = MCPServerConfig(
        name="filesystem",
        command="npx",
        args=["-y", "@modelcontextprotocol/server-filesystem", "/data"],
    )

    async with MCPToolProvider(config) as provider:
        tools = await provider.list_tools()
        result = await provider.invoke("read_file", {"path": "/data/input.txt"})
        return result

YAML Workflow with MCP

A complete workflow that connects agents to external MCP servers:

openintent: "1.0"

info:
  name: "MCP-Enhanced Research Pipeline"
  description: "Agents use MCP tools for file access and data retrieval"

mcp:
  servers:
    filesystem:
      command: "npx"
      args: ["-y", "@modelcontextprotocol/server-filesystem", "/data"]
      allowed_tools: ["read_file", "list_directory"]
      security:
        audit: true
    openintent-remote:
      command: "npx"
      args: ["-y", "@openintentai/mcp-server"]
      env:
        OPENINTENT_SERVER_URL: "http://localhost:8000"
        OPENINTENT_API_KEY: "${OPENINTENT_API_KEY}"
        OPENINTENT_MCP_ROLE: "reader"
      security:
        tls_required: false
        credential_isolation: true
        audit: true

workflow:
  gather:
    title: "Gather Data"
    assign: researcher
  analyze:
    title: "Analyze Findings"
    assign: analyst
    depends_on: [gather]

agents:
  researcher:
    tools:
      - mcp:filesystem/read_file
      - mcp:filesystem/list_directory
  analyst:
    tools:
      - mcp:openintent-remote/openintent_get_intent
      - mcp:openintent-remote/openintent_get_events

Parse this workflow and create a bridge from the mcp: block:

import yaml
from openintent.mcp import MCPBridge

with open("mcp_workflow.yaml") as f:
    workflow = yaml.safe_load(f)

bridge = MCPBridge.from_yaml(workflow["mcp"])
await bridge.connect_all()

tools = await bridge.list_all_tools()
for server, server_tools in tools.items():
    print(f"{server}: {[t['name'] for t in server_tools]}")

result = await bridge.invoke("filesystem", "read_file", {"path": "/data/report.csv"})

await bridge.disconnect_all()

Secure Production Setup

A production configuration with TLS enforcement, role-based access control, tool allowlists, and audit logging.

MCP Server Config (JSON)

{
  "server": {
    "url": "https://openintent.example.com",
    "api_key": "prod-api-key",
    "agent_id": "prod-mcp-agent"
  },
  "security": {
    "tls_required": true,
    "role": "operator",
    "allowed_tools": [
      "openintent_get_intent",
      "openintent_list_intents",
      "openintent_get_events",
      "openintent_update_state",
      "openintent_log_event"
    ],
    "max_timeout": 60,
    "audit_logging": true
  },
  "network": {
    "timeout": 15000,
    "retries": 2,
    "retry_delay": 500
  }
}

Claude Desktop Config (Production)

{
  "mcpServers": {
    "openintent": {
      "command": "npx",
      "args": ["-y", "@openintentai/mcp-server"],
      "env": {
        "OPENINTENT_SERVER_URL": "https://openintent.example.com",
        "OPENINTENT_API_KEY": "prod-api-key",
        "OPENINTENT_MCP_ROLE": "operator",
        "OPENINTENT_MCP_CONFIG": "/etc/openintent/mcp-config.json"
      }
    }
  }
}

Python Bridge (Production)

from openintent.mcp import MCPServerConfig, MCPSecurityConfig, MCPBridge

production_config = MCPServerConfig(
    name="openintent-prod",
    command="npx",
    args=["-y", "@openintentai/mcp-server"],
    env={
        "OPENINTENT_SERVER_URL": "https://openintent.example.com",
        "OPENINTENT_API_KEY": "${OPENINTENT_API_KEY}",
        "OPENINTENT_MCP_ROLE": "operator",
    },
    timeout=15,
    allowed_tools=[
        "openintent_get_intent",
        "openintent_list_intents",
        "openintent_get_events",
    ],
    security=MCPSecurityConfig(
        tls_required=True,
        credential_isolation=True,
        audit=True,
        max_retries=2,
    ),
)

bridge = MCPBridge()
bridge.add_server(production_config)

The security layer validates configuration at startup and warns about potential issues:

  • Missing API keys
  • TLS disabled on non-localhost URLs
  • Excessive timeout values (above 300s)
  • Unknown role values (falls back to reader)
  • admin role in use (flags for operator awareness)

All tool invocations are logged with sanitized parameters — sensitive fields like api_key, password, secret, and token are replaced with [REDACTED] in audit output.