asyncio and uuid are Python standard library modules (available since
Python 3.4+). Listing them in requirements.txt is unnecessary and can
install unrelated PyPI packages with the same name, causing conflicts.
Affected files (8 total):
- starter_ai_agents: openai_research_agent, mixture_of_agents,
ai_meme_generator_agent_browseruse
- mcp_ai_agents: github_mcp_agent, browser_mcp_agent
- ai_agent_framework_crash_course: 5_1_in_memory_conversation_agent,
5_2_persistent_conversation_agent
- advanced_ai_agents: ai_financial_coach_agent
Issue #329 reported that it's unclear how to point browser_mcp_agent at
a local Ollama backend. The mcp-agent framework already supports this
(Ollama exposes an OpenAI-compatible endpoint at
http://localhost:11434/v1), so the gap is documentation + one blocking
env-var check in main.py.
Changes:
- main.py: the hardcoded `os.getenv("OPENAI_API_KEY")` guard
unconditionally rejected users who had correctly configured credentials
via mcp_agent.secrets.yaml (both OpenAI and Ollama cases). Replace
with a check that also accepts a present secrets.yaml, and update the
error to point users at either path.
- mcp_agent.config.yaml: add a commented example showing the `openai:`
block rewritten for Ollama (base_url + default_model). No behaviour
change for existing OpenAI users.
- mcp_agent.secrets.yaml.example: add a commented Ollama example noting
that any non-empty api_key works (Ollama doesn't authenticate).
- README.md: add a "Running with a local Ollama model" section, and
reconcile the previously contradictory "export OPENAI_API_KEY" vs
"use secrets.yaml" instructions into a single 'pick one' step.
Refs #329
Made-with: Cursor
- Spawn MCP servers via StdioServerParameters + stdio_client
- Call list_tools() and convert to Anthropic tool format
- Pass tools to messages.create() calls
- Handle tool_use responses in an agentic loop
- Map tool names to sessions for correct dispatch
Previously agents just used Claude with different system prompts.
Now each agent actually connects to its assigned MCP servers
(github, filesystem, fetch) and can use their tools.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Changed `add_datetime_to_instructions` to `add_datetime_to_context` in travel planner for better context handling.
- Updated response type to `RunOutput` in both travel planner and GitHub agent for improved type safety.
- Specified minimum version for the 'agno' package in requirements.txt for compatibility.
- Replaced Node.js MCP server with the official GitHub MCP server running in Docker.
- Added OpenAI API key input for enhanced query processing.
- Updated README to reflect new requirements and usage instructions.
- Introduced a new AI Real Estate Agent Team with multi-agent capabilities for property search, market analysis, and valuation.
- Added comprehensive README documentation detailing features, setup instructions, and API requirements.
- Implemented the Enterprise MCP AI Agent Team for orchestrating knowledge management across local files and SaaS platforms, including specialized agents for Notion, GitHub, and Figma.
- Enhanced the overall architecture with intelligent routing and task delegation using the Model Context Protocol (MCP).
- Included necessary requirements for both teams to ensure proper functionality.
- Puppeteer puts the entire html of the page into context. Switching to a smaller webpage for demo purposes
- Given the option for the user to switch use_history to false to remove context from calls
George Pickett