Open Source & Technical Projects
Agent infrastructure: the control planes, identity systems, and protocol plumbing that let AI agents do real work.
The open-source server for running fleets of sandboxed coding agents, and the engine under Managoat, its hosted edition. Running Claude instances with worktrees and hand-shuffled MCP configs was slow and error-prone, so I built the tool.
- Spin up sandboxed agent instances with preconfigured env vars, MCP servers, skills, repos, and packages
- Every feature on three surfaces, a web UI for debugging, a REST API for CI/CD, and a Homebrew-distributed CLI with manifest-driven
fountain apply
- Agents onboard themselves, since every instance serves llms.txt and a drop-in skill, so any agentic IDE learns the whole API from one fetch
- The fleet itself is code, and agent-specs declares every agent, environment, and vault as typed resources, applied in one command
- Answers ACP, AG-UI, MCP in both directions, and an OpenAI-compatible endpoint, so clients people already use can drive the fleet
- Streams every conversation over SSE, so agent runs are observable and debuggable rather than fire-and-forget
Built agent-first, in the open. BinaryBourbon is the GitHub identity my agent fleet ships under, and 400+ PRs landed in the first three months, all agent-written. I set the roadmap, review the work, and operate the system, and Fountain is itself developed by the agents it orchestrates.
A production-grade home cloud: three M4 Mac minis running k3s over a Tailscale mesh, every manifest reconciled from git by Flux.
- Full platform stack: Longhorn, Garage, CloudNativePG, Prometheus/Grafana/Loki, SOPS-encrypted secrets
- Serves real production traffic — ai.jakegaylor.com and the agent fleet’s infrastructure run on it
- Photo tour, network topology, and the war stories behind the design
A self-hosted Infisical deployment you drive from the repo, for any Kubernetes cluster kubectl can reach. The shippable artifact behind my secrets-management writing.
- Five components, three tiers (light / production / production-ha), generated CI, and a naming scheme that lets many instances coexist across clusters
- Agent-operable by design: clone it and your agent already knows how to run it by its verbs
- Built with chant, a friend’s intent-aware infrastructure toolchain I use across my stack
A live, hosted MCP server that teaches AI assistants about me. Try it on this candidate.
- Connect any MCP client and it can read my resume, score my fit against your job description, and build the interview
- Your assistant can even email me directly from the conversation — you review it, the server sends it
- Streamable HTTP and SSE transports, an npm package for stdio-only clients, and llms.txt for everyone else
A spec I co-authored: a short list of properties that make infrastructure safe to hand off, to a new engineer or to an AI agent.
- Properties like one path to prod, documentation is law, and named secrets with least privilege, each written to be audited against
- The judgment stays with a human; the rest becomes safe to delegate
- It’s the lens behind this site’s headline, and the standard I hold my own systems to
A free, open service for shareable checklists. Every list is a UUID and knowing the link is the only key — no account required.
- Programmatic by design: a small JSON REST API to create and drive lists, plus a generated llms.txt so an assistant learns the whole API from one fetch
- A built-in, stateless MCP server at
/mcp lets AI assistants create lists, toggle items, and clear checked ones directly
- Optional GitHub sign-in names and tracks your own lists on a “My lists” dashboard; the session-less API and MCP only ever touch open, anonymous lists
- Node/Express + SQLite in one container — deployed to the k3s homelab via Flux GitOps, multi-arch image built in CI, TLS via cert-manager, built agent-first with Claude Code
A hobby project my wife and I are building together: it pairs novelists with critique partners, then has them swap chapters a few thousand words at a time.
- Matches on mutual fit, ranking writers for whom the trade works in both directions
- Pages burn after fourteen days, so drafts stay private; the notes you received are yours forever
- Live now, and my wife is its first and toughest user
Early participant in Model Context Protocol (MCP) community development.
- Was a member of the MCP Hosting Working Group
- Developed strategies for deploying and running MCP servers at scale
- Contributed to the Dart SDK and an Express tool for binding to MCP servers
- Built MCP servers for Hirebase, candidate job search, and text extraction
- Created a GitOps tool for MCP deployments; still operate a public Echo MCP server for community testing
CareerFlick
Flutter mobile app with a Tinder-like LLM interface.
- Built an intuitive swipe-based interface for interacting with LLMs using Flutter
- Implemented AI-driven content generation and response system
- Shipped cross-platform for iOS and Android
An MCP server for the CleanJobData Job API, published on PyPI in August 2026.
- Runs as stdio for local clients or as one hosted HTTP server where every user authenticates with their own API key per request
- The newest entry in a line of MCP servers I’ve shipped since the protocol’s early days
A Python MCP server for Hirebase.org’s API.
- Built a Python-based MCP server for searching jobs on Hirebase.org
- Dockerized the service and set up GitOps CI/CD pipelines for automated builds and releases
Package to simplify MCP handling in Express.js.
- Developed Express middleware in TypeScript to streamline MCP request handling
- Authored full type definitions and robust error handling for production-grade reliability
A TypeScript library to build customizable candidate MCP servers.
- Authored a reusable library so developers can integrate their own candidate data into MCP flows
- Published as an npm package for easy adoption and extension by other teams
Dart SDK implementation of the Model Context Protocol.
- Contributed stream-based support for in-process MCP servers
- Added support for running MCP servers in Flutter apps
- Authored Dart examples to accelerate SDK adoption