arifOS MCP
arifOS evaluates actions an agent proposes against 13 rules before execution and returns a verdict with a signed receipt
- 53 ★ on the repository
- Last commit
Extended access
Why this level
- Agent proposals and their context go to a remote server unless a local copy is run
- A verdict does not replace an independent safety review
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/arifosDetects 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.
- Agent proposals and their context go to a remote server unless a local copy is run
- A verdict does not replace an independent safety review
This is third-party code. Review the repository files before installing.
What it does
arifOS is a model-independent decision layer above AI agents. It checks an agent's proposal against thirteen constitutional rules before execution, returns a verdict, keeps the human veto and writes a receipt to the VAULT999 store. Execution stays with a separate component, and arifOS itself is neither a model nor an agent framework. Over MCP eight canonical verbs are available: `arif_init`, `arif_observe`, `arif_think`, `arif_route`, `arif_memory`, `arif_judge`, `arif_forge` and `arif_seal`. The public address `https://mcp.arif-fazil.com/mcp` works over Streamable HTTP, your own copy can run locally at `http://localhost:8088/mcp`, and the package is `arifos` on PyPI. The author, an exploration geoscientist, built the system for decisions under incomplete observations.
Who it is for. Researchers and teams building agents with decision checks before execution who want audit receipts.
Good fit when
- An agent puts forward a dangerous action and the kernel issues a verdict before execution
- Every decision is stored as a receipt for later checking
- Your own copy of the kernel has to run locally on port 8088
- A decision must keep the human's right of veto
Not a fit when
- A ready executor of actions is needed, and arifOS does not contain one
- An independently verified safety guarantee is required
- An agent's proposals must not go to an external address
Example request
Evaluate the proposed action to delete old backups on production and return a verdict with a receiptMCP
- Transport
- http
- Authentication
- not required
Limitations
The public endpoint works over Streamable HTTP at `https://mcp.arif-fazil.com/mcp` and agent proposals go to that address, while a local copy at `http://localhost:8088/mcp` serves offline work. The README itself shows a discrepancy: the kernel's own measurements mark 12 of 13 rules as passing, while an independent audit of 11 October 2026 found failed and unmeasured rules, so this cannot be counted as a safeguard. Execution of decisions is done by a separate component, A-FORGE. Documentation is in English only.
How to disable. Remove the server entry from the client configuration. Stop a local copy and uninstall the package with `pip uninstall arifos`.
FAQ
How do I connect arifOS?
Add the Streamable HTTP remote server `https://mcp.arif-fazil.com/mcp` or install the `pip install arifos` package and start a local copy at `http://localhost:8088/mcp`.
Which tools are available?
Eight verbs: `arif_init`, `arif_observe`, `arif_think`, `arif_route`, `arif_memory`, `arif_judge`, `arif_forge`, `arif_seal`. The schemas are published on the kernel site.
What are the 13 rules?
Thirteen constitutional criteria by which the kernel judges a proposal before execution. The state of each measurement is shown in the README and refreshed automatically.
Does arifOS execute actions?
No, execution belongs to a separate component, A-FORGE. arifOS issues a verdict, keeps the human veto and writes a receipt.
Can the result be trusted as a safety guarantee?
The README itself separates self-measurement from independent audit and notes that unmeasured is not pass. Treat a verdict as an extra signal.
Discussion
No questions or reviews yet.
Related
MCP serversMCP Agent Mail
mcp_agent_mail
A mail coordination layer for coding agents: agents have memorable names, inboxes and outboxes, history search and voluntary file reservations
A terminal agent and Python framework with MCP, SKILL.md and ACP support for coding, building and evaluating your own agents
A mail server for agents sharing one repository: messages, threads, file reservations and a 16-screen TUI
An MCP server that runs sub-agents defined in Markdown through Codex, Claude Code, Cursor CLI and other CLIs from any MCP client
A universal MCP CLI client from Apify: persistent sessions, OAuth, JSON output for scripts and shell access to any MCP server for agents
Local agent memory built on a corrections ledger: tell Claude Code once and a new session reads the rule before it starts work
README in short
The README opens with a block of the current verified state that separates self-measurement from independent audit. It then sets out the idea of authority before execution, the architecture with VAULT999 and A-FORGE, a quick start and a section on what is not yet proven.