[How-To] Agentic Workflows: From Local OpenClaw to External MCP Hives
Feb 5 Office Hours Recap: A technical guide to running Telegram bots locally, automating email triage, and porting Brain Trust tools to your IDE
The February 5th session covered four specific workflows: custom email agents, local agent orchestration via Telegram, event-driven calendar triggers, and using MCP Hives externally.
1. The “Gatekeeper”: Custom Email Orchestration
While tools like Shortwave offer packaged email management, they often add noise. David Proctor demonstrated “Gatekeeper,” a custom Brain Trust agent that offers superior control through natural language instructions.
The Workflow:
Instead of a generic filter, the agent was explicitly instructed to:
Identify and forward receipts directly to Expensify.
Archive newsletters and marketing spam.
Flag items requiring specific human action.
Safety Check: The agent was deliberately denied “delete” permissions, opting to archive instead.
The Gateway Lesson:
The value lies in orchestration. A single agent can manage the inbox, but it can also be called upon to summarize threads and draft updates for the team.
Resources:
Brain Trust Experts: http://mcp-hive.ti.trilogy.com/experts/
Gatekeeper Agent: https://mcp-hive.ti.trilogy.com/experts/gatekeeper
2. OpenClaw & Telegram: Remote Control for Local Agents
Stan Huseletov introduced OpenClaw, an open-source tool that allows AI agents to run locally on your machine, accessing your file system and code repositories. He demonstrated how to bridge this local power with a mobile interface using Telegram.
Implementation Steps:
Bot Creation: Use “Botfather” on Telegram to generate a new bot and API token.
Installation: Install OpenClaw locally via the command line.
Integration: Input the Telegram token and an Anthropic API key into the OpenClaw configuration.
Security & Utility:
This turns Telegram into a command line for your local machine. You can ask the bot to debug code or check system health remotely.
Note: As discussed in the chat, connecting to G-Suite/Drive files carries risk. It is recommended to run this in a Virtual Machine or restricted folder.
Resources & Commands:
Official Site (Install Command): https://openclaw.ai/
Security & Analysis (LinkedIn Post): Trilogy AI CoE OpenClaw Analysis
3. Automating the Automation: Calendar Triggers
David showcased a method to make agents proactive rather than reactive using Calendar Triggers.
The Mechanism:
An agent (in this case, “The Algorithm”) monitors a Google Calendar. When it detects a specific event title (e.g., “The Algorithm”), it executes a pre-defined policy.
Business Use Case:
You can set an agent to trigger 30 minutes before a recurring “Weekly Business Review.” The agent will pull metrics and generate a report exactly when you need it, regardless of when the meeting is scheduled.
Resources:
The Algorithm Live: https://thealgorithm.live/
The Algorithm Expert: https://mcp-hive.ti.trilogy.com/experts/thealgorithm
4. Portable Tools: Using Hives Externally
The session concluded with a technical look at the Model Context Protocol (MCP). In Brain Trust, “Hives” act as MCP servers housing specific tools.
The Feature:
These tools are not locked to Brain Trust. Users can copy the Connection URL of a Hive and plug it directly into external environments like Cursor, Windsurf, or Claude Desktop. This allows for a “build once, use everywhere” architecture.
Resources:
MCP Hive Home: https://mcp-hive.ti.trilogy.com/home
5. Recommended Learning: Learners Lens
To dive deeper into these topics, the team has curated specific courses on Learners Lens. These cover practical workflows mentioned in the Office Hours, including safe Claude deployment and building MCP bots.
Claude Code Production Workflows, Safely
Supercharge Workflows with AI Agents
Writing Reliable Skill Files for Agents
Build and Launch Viral MCP Bots
These workflows make agentic automation tangible, showing how local control, proactive triggers, and portable MCP tools can really streamline complex tasks