# AutoGen (/docs/frameworks/autogen) Give an AutoGen agent social media data, with the price shown before it spends. AutoGen can attach an MCP server to an agent through `McpWorkbench`, so the hosted SocialCrawl server gives an agent its tools with no function to write. SocialCrawl publishes no AutoGen package, and the one install line below is AutoGen's own. ## Setup Install AutoGen with its MCP and OpenAI extras, then set your keys. Create the SocialCrawl key in the dashboard under API Keys; it looks like `sc_your_key_here`. ```bash title="Terminal" pip install -U autogen-agentchat "autogen-ext[mcp,openai]" export SOCIALCRAWL_API_KEY=sc_your_key_here export OPENAI_API_KEY=your_openai_key ``` ## Connect Point a workbench at the hosted server at `https://mcp.socialcrawl.dev/mcp` and send your key as a bearer header. The server brings a free estimate tool, `socialcrawl_estimate`, so the system message can ask for the price first. `max_tool_iterations` is the step limit. ```python title="Python" import asyncio import os from autogen_agentchat.agents import AssistantAgent from autogen_ext.models.openai import OpenAIChatCompletionClient from autogen_ext.tools.mcp import McpWorkbench, StreamableHttpServerParams server = StreamableHttpServerParams( url="https://mcp.socialcrawl.dev/mcp", headers={"Authorization": f"Bearer {os.environ['SOCIALCRAWL_API_KEY']}"}, ) SYSTEM = ( "Read public social media data with SocialCrawl. Call socialcrawl_estimate and report the cost " "before any paid call. If a call says the account is out of credits, tell the user and stop." ) async def main() -> None: model_client = OpenAIChatCompletionClient(model="gpt-4.1-mini") async with McpWorkbench(server) as workbench: agent = AssistantAgent( "researcher", model_client=model_client, workbench=workbench, system_message=SYSTEM, reflect_on_tool_use=True, max_tool_iterations=5, ) result = await agent.run(task="Read the TikTok profile for nasa and summarise it.") print(result.messages[-1].content) await model_client.close() asyncio.run(main()) ``` ## Read a profile and its posts The run above asks the agent to quote before it reads, so the estimate appears in the tool calls that `result.messages` records. Widen the task to ask for the latest videos as well, for example `Read the TikTok profile for nasa and its latest videos, then summarise both.` ## Keep spend in check Call `estimate_cost` before any paid call and show the answer to the person who owns the key. `max_tool_iterations` caps the loop, and `max_credits` stops a list walk where you set it. A 402 means the account is out of credits. Tell the user it is an offer to top up and stop, because a retry cannot succeed. See [Errors](/docs/errors.md) and [Credits](/docs/credits.md). ## Researcher and writer Put a second agent in a round-robin team. Only the researcher gets the workbench, so only the researcher can spend credits, and the writer ends the run by saying `DONE`. ```python title="Python" from autogen_agentchat.conditions import TextMentionTermination from autogen_agentchat.teams import RoundRobinGroupChat async def team_run() -> None: model_client = OpenAIChatCompletionClient(model="gpt-4.1-mini") async with McpWorkbench(server) as workbench: researcher = AssistantAgent( "researcher", model_client=model_client, workbench=workbench, system_message=SYSTEM, reflect_on_tool_use=True, max_tool_iterations=5, ) writer = AssistantAgent( "writer", model_client=model_client, system_message="Write a four-sentence brief from the researcher's notes, then say DONE.", ) team = RoundRobinGroupChat([researcher, writer], termination_condition=TextMentionTermination("DONE")) result = await team.run(task="Read the TikTok profile for nasa and brief me on it.") print(result.messages[-1].content) await model_client.close() asyncio.run(team_run()) ``` ## Resources - [AI agents](/docs/ai-agents.md): Machine-readable references and agent tool files. - [Skills and MCP](/docs/skills-and-mcp.md): The hosted server and agent skill, with no wrapper to write. - [Errors](/docs/errors.md): What each error type means and which are safe to retry. - [Endpoint pricing](/docs/endpoint-pricing.md): The credit price of every endpoint.