Problem

When creating a cron job that uses skills or relies on MCP servers, the job fails because the cron scheduler runs in a separate Python environment that does not share the MCP/plugin configuration loaded by the main CLI process.

Root Cause

In cron/scheduler.py, the run_job() function creates an AIAgent without initializing MCP servers. The MCP connections are established in cli.py when the main process starts, but cron jobs execute in a context where discover_mcp_tools() has never been called.

Key issues in cron/scheduler.py:

1. MCP not initialized: No call to discover_mcp_tools() before creating the AIAgent
2. Skills work differently: Skills are loaded via _build_job_prompt() which reads skill files directly - this works, but the skill execution (e.g., using skill tools) may still fail if the skill toolset is not properly initialized
3. Plugins: Similarly, plugin configuration is not propagated to cron job runs

Expected Behavior

Cron jobs should be able to use:

• Skills (defined via skill: or skills: in the job)
• MCP servers (defined in config.yaml under mcp_servers)
• Any plugins configured in the main hermes config

Environment

• Hermes version: latest
• Python environment: separate cron process
• MCP servers: configured in ~/.hermes/config.yaml

Possible Solution

In run_job(), after loading config.yaml, initialize MCP before creating the AIAgent:

# Initialize MCP servers for this cron job run
try:
    from tools.mcp_tool import discover_mcp_tools, _ensure_mcp_loop
    _ensure_mcp_loop()
    mcp_tools = discover_mcp_tools()
    logger.info("Job '%s': discovered %d MCP tools", job_id, len(mcp_tools))
except Exception as e:
    logger.warning("Job '%s': failed to initialize MCP: %s", job_id, e)