Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -2,8 +2,9 @@

**Paper:** [Tracking Capabilities for Safer Agents (arXiv:2603.00991)](https://arxiv.org/abs/2603.00991)

TACIT (Tracked Agent Capabilities In Types) is an [MCP](https://modelcontextprotocol.io/) server that provides a **safety harness** for AI agents.
TACIT (Tracked Agent Capabilities In Types) is a **safety harness** for AI agents.
Instead of calling tools directly, agents write code in Scala 3 with [capture checking](https://nightly.scala-lang.org/docs/reference/experimental/capture-checking/index.html): a type system that statically tracks capabilities and enforces that agent code cannot forge access rights, cannot perform effects beyond its budget, and cannot leak information from pure sub-computations.
It provides an [MCP](https://modelcontextprotocol.io/) interface, so that it can be easily used by all MCP-compatible agents.

![TACIT Framework Overview](diagrams/overall.png)

Expand All @@ -15,7 +16,7 @@ The framework has three main components:

## Quick Start

TACIT is a standard MCP server that communicates via JSON-RPC over stdio. It works with any MCP-compatible agent, including Claude Code, OpenCode, GitHub Copilot, and others.
TACIT provides a standard MCP server that communicates via JSON-RPC over stdio. It works with any MCP-compatible agent, including Claude Code, OpenCode, GitHub Copilot, and others.

### 1. Build

Expand Down
Loading