Rbbt’s workflow engine (Scout Workflow) can be exposed to LLMs as callable tools.

In the Scout ecosystem, this is provided by Scout-AI:

  • LLM.ask / LLM::Agent for stateful conversations
  • a chat-file format with special roles like tool: and mcp:
  • automatic execution of tool calls (workflows / knowledge bases / MCP servers)

This page is intentionally practical: a minimal working pattern you can copy.

Prerequisites

You need an environment with Scout-AI available (CLI: scout-ai).

If you are using a full Rbbt distribution that bundles Scout-AI, the commands below should work. Otherwise, install/configure Scout-AI separately.

Pattern A: expose a workflow as tools in a chat file

Create a file baking.chat:

user:

# (optional but recommended) inject workflow docs so the model knows what tools do
introduce: Baking

# expose all tasks of the workflow as callable tools
tool: Baking

Bake muffins using the workflow tools. Prefer calling the tool instead of writing steps.

Run it:

scout-ai llm ask -c baking.chat -e nano

What happens:

  1. The chat is compiled (imports, tool declarations, file inclusion).
  2. The model can call workflow task functions.
  3. Tool calls are executed and the trace is appended to the chat.

Pattern B: agent CLI (workflow-aware)

Scout-AI also provides an agent wrapper that automatically exports workflow tasks as tools:

scout-ai agent ask Baking "Bake muffins using the tool"

MCP integration (optional)

If you already have tools exposed via an MCP server, chat files can load them too:

mcp: http://localhost:8765

Or using an MCP stdio server:

mcp: stdio 'npx -y @modelcontextprotocol/server-filesystem ${pwd}'

You can combine MCP tools and workflow tools in the same chat.

Where to read the full reference

  • Scout-AI chat roles (tool, introduce, task, mcp, …)
  • How tool calling is executed and how endpoints/models are configured

See:

This site uses Just the Docs, a documentation theme for Jekyll.