# AI Agents & Workflow Platforms

> Built production-oriented agent systems and the supporting tools, contracts, authorization boundaries and operational infrastructure needed to use them safely.

- **Role:** Agent, MCP & Automation Engineering
- **Category:** Professional work
- **Human-readable case study:** https://lutherminshull.com/projects/ai-agents-internal-tooling

## Context

The useful part of agent engineering is often the system around the model: reliable access to business context, typed workflows, review boundaries, auditable artifacts and repeatable deployment.

## Luther's role

- Built agent workflows using Eve and OpenClaw across internal and client operations.
- Designed and shipped the complete ActiveCollab integration from API client through CLI and MCP transports.
- Turned an inherited quoting prototype into a production-oriented agent workflow with explicit job and artifact contracts.
- Defined read-only access, human-review states, environment isolation and operational deployment boundaries.

## What Luther built

- Authenticated ActiveCollab tools for projects, tasks, comments, attachments, users, companies and time records.
- CLI, stdio MCP and HTTP MCP interfaces over a shared integration layer, with installer and distribution workflows.
- Typed agent job, event, artifact and result contracts with deterministic local test workers.
- Auditable web and email intake, provider-independent storage and synchronized agent outputs.
- Pricing-mirror integration, mismatch flags and explicit needs-review states for AI-assisted quoting.
- Security hardening, continuous integration and deployment automation for production agent services.

## How it works

- Read-only MCP capabilities built around explicit authentication and operational boundaries.
- Structured worker contracts that normalize model output before it becomes application state.
- Human attention represented as a first-class workflow state rather than an unstructured model response.
- Agent processes isolated from database and storage credentials through narrow, allowlisted interfaces.

## Outcomes

- Made project context available to team AI assistants through authenticated, repeatable tooling.
- Created reusable foundations for production agent workflows beyond a single chat interface.
- Connected AI-assisted work to observable jobs, review states and downloadable artifacts.

## Technologies

TypeScript, MCP, Eve, OpenClaw, Next.js, Drizzle, PostgreSQL, Docker, CI/CD

## Contact

- [Discuss an opportunity](https://cal.com/lutherminshull)
- [Email Luther](mailto:luther@lutherminshull.com)
- [Return to the portfolio overview](https://lutherminshull.com/index.md)

## Scope note

This case study is limited to public and non-sensitive implementation details. Client-confidential information and internal data are not included.