Hexis: a catalogue of skills and context for AI agents
Hexis
A self-hosted git store for skills, tools and context: agents fetch them over MCP and every change goes through review
- 125 ★ on the repository
- Last commit
Extended access
Why this level
- The agent receives company instructions and context: a tampered skill in the repository changes the behavior of every agent, so review and protected branches matter
- The deployment holds integration secrets and SSO sign-in: keep `JWT_SECRET` and `SECRETS_ENC_KEY` safe
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/bevel-software-hexisDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent receives company instructions and context: a tampered skill in the repository changes the behavior of every agent, so review and protected branches matter
- The deployment holds integration secrets and SSO sign-in: keep `JWT_SECRET` and `SECRETS_ENC_KEY` safe
This is third-party code. Review the repository files before installing.
What it does
Hexis keeps a company's skills, tool manuals and reference material as files in your own git repository. Agents connect over MCP and read what their user's role allows. The tools `list_skills`, `search` and `get_skill` let an agent narrow the field first and then load a single skill, so context does not bloat. Changes go through change requests. Any member or the agent itself can propose a new skill or a fix, and on protected branches it reaches the owners of the files it touches and takes effect once they approve. Every save is a commit, so history, blame and revert work as they do for code. The app runs from a Docker Compose image with PostgreSQL, with sign-in set up through SSO. The agent never holds the credentials it uses: access is resolved per file. You can connect to a hosted server over Streamable HTTP with OAuth or start a local server with `npx -y @bevel-software/hexis-mcp`.
Who it is for. Teams and IT departments that want to keep shared skills, instructions and access rights for agents in one repository on their own servers, with change review.
Good fit when
- Agent skills sit in different personal folders and need one place with history
- A skill edit must be approved by its owner, not by any member
- Skill access is granted to roles, and each role gets its own set
- Non-technical colleagues should propose and browse skills without working in git
- A company wants to run a knowledge base for agents on its own infrastructure behind its own SSO
Not a fit when
- You work alone with few skills: a plain skills folder is enough, no server, PostgreSQL or SSO
- You only need a gateway to ready-made MCP servers without your own rules and review: MCP gateways cover that
- You have no server or Docker: the app is meant to be self-hosted
Example request
Find the sales deck skill in Hexis, load it and build slides for a new client with it.MCP
- Transport
- stdio, http
- Authentication
- OAuth
Limitations
Running it takes a server with Docker Compose, an empty git repository for the knowledge base and four required `.env` values: the admin email, a password, `JWT_SECRET` and `SECRETS_ENC_KEY`.
A public HTTPS address needs a domain and open ports 80 and 443. The author advises pinning the image version in `.env` so an upgrade never arrives unplanned. The public demo is shared and mostly read-only: visitors propose changes and owners approve them.
The local MCP server requires Node 22.13 or 24. The author offers a managed instance separately, by email.
How to disable. Remove the server in the MCP client. The deployment is stopped with `docker compose down`, and the knowledge base stays an ordinary git repository.
FAQ
How is Hexis different from MCP gateways?
A gateway hands ready tools to users in one direction. In Hexis any member or agent proposes a change, the file owners approve it, and everything is stored in git with history.
Which agents can connect?
Any MCP-capable client, including Claude Code, Codex, Cursor, Cline and ChatGPT. Each sees only what its user's role allows.
How does an agent find a skill without flooding context?
The agent calls `list_skills` and `search` to narrow the field, then `get_skill` for one skill. A description of the knowledge base is passed at the start of every MCP session from `mcp-description.md` in the repository root.
Can I try it without installing?
Yes, there is a public demo at demo.bevel.software with a fictional company. Sign in with a Google account, and the Start here page walks through the whole loop: connecting an agent, using a skill and receiving a proposed change.
How do I deploy my own copy?
Download `docker-compose.yml` and `.env.example` from a release, fill the four required variables and run `docker compose up -d`. For public HTTPS add the `https` profile with Caddy.
Discussion
No questions or reviews yet.
Related
MCP serversMeta Skill (ms)
Meta Skill
Meta Skill (ms) keeps skills locally in SQLite and Git, searches them with a hybrid index and serves agents over MCP
A skill registry for teams with security scanning and versioning; the MCP server finds and installs skills on request.
A Rust app for macOS, Linux and Windows: skills, MCP servers, plugins, hooks and configs of 17 AI agents in one window, with a security audit
A Codex skill collection: phased planning, parallel subagents, documentation access, a browser, 51 hooks and ready agent roles.
A local MCP server that stores skills as versions and lets improved variants go live only after a human review
A CLI and MCP server for Agent Skills: spec validation, installs from GitHub and on-demand loading of the right skill
README in short
The project is positioned as the open core of the Bevel platform and a git-backed control plane for agent skills, tools and permissions. The README covers a Docker quick start, connecting Cline, a configuration reference and an FAQ on versioning and governance.