feat: add langgraph with MCP example

acp/ollama_weather_service/Dockerfile

@@ -0,0 +1,9 @@
+FROM python:3.11-slim-bookworm
+ARG RELEASE_VERSION="main"
+COPY --from=ghcr.io/astral-sh/uv:0.6.16 /uv /bin/
+WORKDIR /app
+COPY . .
+RUN uv sync --no-cache --locked --link-mode copy
+ENV PRODUCTION_MODE=True \
+    RELEASE_VERSION=${RELEASE_VERSION}
+CMD ["uv", "run", "--no-sync", "server"]

acp/ollama_weather_service/pyproject.toml

@@ -0,0 +1,31 @@
+[project]
+name = "ollama-weather-service"
+version = "0.0.1"
+description = "Simple ollama-based Langgraph agent with MCP tool calling."
+authors = [
+    { name = "Paolo Dettori" }
+]
+readme = "README.md"
+license = { text = "Apache" }
+requires-python = ">=3.11"
+dependencies = [
+    "acp-sdk>=0.8.1",
+    "langgraph>=0.2.55",
+    "langchain-community>=0.3.9",
+    "tavily-python>=0.5.0",
+    "langchain-ollama>=0.2.1",
+    "duckduckgo-search~=7.5.5",
+    "beautifulsoup4>=4.13.3",
+    "langchain-openai>=0.3.7",
+    "openinference-instrumentation-langchain>=0.1.36",
+    "pydantic-settings>=2.8.1",
+    "langchain-mcp-adapters>=0.0.10",
+]
+
+[project.scripts]
+server = "ollama_weather_service.agent:run"
+
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"

acp/ollama_weather_service/src/ollama_weather_service/.dockerignore

@@ -0,0 +1 @@
+.venv

acp/ollama_weather_service/src/ollama_weather_service/agent.py

@@ -0,0 +1,100 @@
+import os
+from textwrap import dedent
+from typing import AsyncIterator
+
+from acp_sdk import Metadata, Message, Link, LinkType, MessagePart
+from acp_sdk.server import Server
+from openinference.instrumentation.langchain import LangChainInstrumentor
+from pydantic import AnyUrl
+from langchain_core.messages import HumanMessage
+
+from ollama_weather_service.graph import get_graph, get_mcpclient
+
+
+LangChainInstrumentor().instrument()
+
+server = Server()
+
+
+@server.agent(
+    metadata=Metadata(
+        programming_language="Python",
+        license="Apache 2.0",
+        framework="LangGraph",
+        links=[
+            Link(
+                type=LinkType.SOURCE_CODE,
+                url=AnyUrl(
+                    f"https://github.com/i-am-bee/beeai-platform/blob/{os.getenv('RELEASE_VERSION', 'main')}"
+                    "/agents/community/ollama-weather-service"
+                ),
+            )
+        ],
+        documentation=dedent(
+            """\
+            This agent provides a simple weather infor assistant.
+
+            ## Input Parameters
+            - **prompt** (string) – the city for which you want to know weather info.
+
+            ## Key Features
+            - **MCP Tool Calling** – uses a MCP tool to get weather info.
+            """,
+        ),
+        use_cases=[
+            "**Weather Assistant** – Personalized assistant for weather info.",
+        ],
+        env=[
+            {"name": "LLM_MODEL", "description": "Model to use from the specified OpenAI-compatible API."},
+            {"name": "LLM_API_BASE", "description": "Base URL for OpenAI-compatible API endpoint"},
+            {"name": "LLM_API_KEY", "description": "API key for OpenAI-compatible API endpoint"},
+            {"name": "MCP_URL", "description": "MCP Server URL for the weather tool"},
+        ],
+        ui={"type": "hands-off", "user_greeting": "Ask me about the weather"},
+        examples={
+            "cli": [
+                {
+                    "command": 'beeai run ollama_weather_service "what is the weather in NY?"',
+                    "description": "Running a Weather Query",
+                    "processing_steps": [
+                        "Calls the weather MCP tool to get the weather info"
+                        "Parses results and return it",
+                    ],
+                }
+            ]
+        },
+    )
+)
+async def ollama_weather_service(input: list[Message]) -> AsyncIterator:
+    """
+    The agent allows to retrieve weather info through a natural language conversatinal interface
+    """
+    messages = [HumanMessage(content=input[-1].parts[-1].content)]
+    input = {"messages": messages}
+    print(f"{input}")
+    try:
+        output = None
+        async with get_mcpclient() as mcpclient:
+            graph = await get_graph(mcpclient)
+            async for event in graph.astream(input, stream_mode="updates"):
+                yield {
+                    "message": "\n".join(
+                        f"🚶‍♂️{key}: {str(value)[:100] + '...' if len(str(value)) > 100 else str(value)}"
+                        for key, value in event.items()
+                    )
+                    + "\n"
+                }
+                output = event
+                print(event)
+            output = output.get("assistant", {}).get("messages")
+            yield MessagePart(content=str(output))
+    except Exception as e:
+        raise Exception(f"An error occurred while running the graph: {e}")
+
+
+def run():
+    server.run(host=os.getenv("HOST", "127.0.0.1"), port=int(os.getenv("PORT", 8000)))
+
+
+if __name__ == "__main__":
+    run()

acp/ollama_weather_service/src/ollama_weather_service/graph.py

@@ -0,0 +1,48 @@
+from langgraph.graph import StateGraph, MessagesState, START
+from langchain_mcp_adapters.client import MultiServerMCPClient
+from langchain_core.messages import SystemMessage
+from langgraph.prebuilt import tools_condition, ToolNode
+from langchain_ollama import ChatOllama
+import os
+
+def get_mcpclient():
+    return MultiServerMCPClient({
+        "math": {
+            "url": os.getenv("MCP_URL", "http://localhost:8000/sse"),
+            "transport": "sse",
+        }
+    }
+    )
+
+async def get_graph(client) -> StateGraph:
+
+    llm = ChatOllama(
+        model="llama3.1",
+        openai_api_key="http://localhost:11434/v1",
+        openai_api_base="dummy",
+        temperature=0,
+    )
+    llm_with_tools = llm.bind_tools(client.get_tools())
+
+    # System message
+    sys_msg = SystemMessage(content="You are a helpful assistant tasked with providing weather information. You must use the provided tools to complete your task.")
+
+    # Node
+    def assistant(state: MessagesState):
+        return {"messages": [llm_with_tools.invoke([sys_msg] + state["messages"])]}
+
+    # Build graph
+    builder = StateGraph(MessagesState)
+    builder.add_node("assistant", assistant)
+    builder.add_node("tools", ToolNode(client.get_tools()))
+    builder.add_edge(START, "assistant")
+    builder.add_conditional_edges(
+        "assistant",
+        tools_condition,
+    )
+    builder.add_edge("tools", "assistant")
+
+    # Compile graph
+    graph = builder.compile()
+    return graph
+