PICA MCP Integration with CrewAI

PICA provides a unified API platform that connects AI agents to third-party services (CRMs, email, calendars, databases, etc.) through MCP tool calling.

PICA MCP Server

PICA exposes its capabilities through an MCP server distributed as @picahq/mcp. It uses stdio transport — it runs as a local subprocess via npx.

MCP Configuration

{
  "mcpServers": {
    "pica": {
      "command": "npx",
      "args": ["@picahq/mcp"],
      "env": {
        "PICA_SECRET": "your-pica-secret-key"
      }
    }
  }
}

• Package: @picahq/mcp (run via npx, no install needed)
• Auth: PICA_SECRET environment variable (obtain from the PICA dashboard https://app.picaos.com/settings/api-keys)
• Transport: stdio (standard input/output)

Environment Variable

Always store the PICA secret in an environment variable, never hardcode it:

PICA_SECRET=sk_test_...

Add it to .env and load with python-dotenv.

Using PICA with CrewAI

CrewAI provides built-in MCP support via MCPServerStdio from crewai.mcp. Always refer to the latest docs before implementing. See crewai-mcp-reference.md.

Required packages

pip install "crewai[anthropic]" python-dotenv

The [anthropic] extra is required when using Anthropic models — without it you get Anthropic native provider not available.

Before implementing: look up the latest docs

The CrewAI API evolves across versions. Always check the latest docs before writing code. See crewai-mcp-reference.md.

Integration pattern

1. Attach MCP servers directly to an Agent via the mcps parameter using MCPServerStdio
2. Configure the LLM with LLM(model="anthropic/claude-haiku-4-5-20251001", stream=True)
3. Create a Task and Crew, then call crew.kickoff()
4. Pass environment variables (PICA_SECRET, PATH, HOME) to MCPServerStdio's env config

Minimal example

from crewai import Agent, Task, Crew, Process, LLM
from crewai.mcp import MCPServerStdio

llm = LLM(model="anthropic/claude-haiku-4-5-20251001", stream=True)

agent = Agent(
    role="PICA Assistant",
    goal="Help the user with their request",
    backstory="You are a helpful AI assistant with access to PICA tools.",
    llm=llm,
    mcps=[
        MCPServerStdio(
            command="npx",
            args=["@picahq/mcp"],
            env={
                "PICA_SECRET": os.environ.get("PICA_SECRET", ""),
                "PATH": os.environ.get("PATH", ""),
                "HOME": os.environ.get("HOME", ""),
            },
        ),
    ],
    verbose=True,
)

task = Task(
    description="User's prompt here",
    expected_output="A helpful response addressing the user's request.",
    agent=agent,
)

crew = Crew(agents=[agent], tasks=[task], process=Process.sequential, verbose=True)
result = crew.kickoff()

Important: streaming with CrewAI events

CrewAI runs synchronously. To stream responses (e.g., via FastAPI SSE), run the crew in a thread executor and use an event listener to forward chunks:

from crewai.events import BaseEventListener, LLMStreamChunkEvent

class StreamListener(BaseEventListener):
    def setup_listeners(self, crewai_event_bus):
        @crewai_event_bus.on(LLMStreamChunkEvent)
        def on_chunk(source, event):
            # event.chunk contains the text chunk
            pass

Thread safety: when using asyncio.Queue to bridge the sync crew thread and an async event loop, use loop.call_soon_threadsafe(queue.put_nowait, item) instead of queue.put_nowait() directly — asyncio.Queue is not thread-safe.

Key event types

• LLMStreamChunkEvent — streaming text chunks (event.chunk)
• ToolUsageStartedEvent — tool call started (event.tool_name, event.tool_input)
• ToolUsageFinishedEvent — tool call finished (event.tool_name, event.tool_output)
• CrewKickoffCompletedEvent — crew finished execution

Checklist

When setting up PICA MCP with CrewAI:

• crewai[anthropic] is installed (the [anthropic] extra is required)
• PICA_SECRET is set in .env
• .env is loaded via python-dotenv (load_dotenv() at top of file)
• MCPServerStdio uses npx @picahq/mcp with PICA_SECRET, PATH, HOME in env
• LLM model string uses anthropic/ prefix (e.g., anthropic/claude-haiku-4-5-20251001)
• If streaming, event listener uses loop.call_soon_threadsafe() for thread-safe queue puts
• Entire crew setup is wrapped in try/except with a finally to signal completion

Additional resources

• For CrewAI MCP docs and API details, see crewai-mcp-reference.md