# Bartholomew Autonomous Trust Protocol (BTP v2.4)
> The transactional MCP security proxy and sub-5µs micro-rollback runtime for autonomous AI agents.

## Overview
Bartholomew provides transparent Model Context Protocol (MCP) proxy interception and transactional workspace rollbacks for autonomous agents interacting with shell environments, filesystems, databases, and APIs.

## Core Capabilities
- **Transactional Micro-Rollbacks (<5 µs)**: In-memory Copy-on-Write byte snapshots restore corrupted files and path escapes in 2.30 µs with zero residual state.
- **Bi-Directional Credential Scrubbing**: Intercepts and redacts OpenAI, Anthropic, AWS, and GitHub credentials in both incoming agent requests and outgoing tool stdout responses.
- **Scoped AST Invariant Analysis**: Scoped symbol-table static analyzer neutralizing dynamic execution evasions (`importlib`, `asyncio.create_subprocess_shell`, `pickle.loads`).
- **Chained Merkle Audit Graph**: Every turn is bound to prior state via RFC 8785 canonical JSON and FIPS Ed25519 digital signatures for offline enterprise compliance.

## Machine Discovery Endpoints
- **Agent Manifest**: `https://acn-fastapi-backend-322603900775.us-central1.run.app/.well-known/ai-agent-manifest.json`
- **BTP Configuration**: `https://acn-fastapi-backend-322603900775.us-central1.run.app/.well-known/btp-configuration.json`
- **Live Sidecar Gateway**: `https://acn-fastapi-backend-322603900775.us-central1.run.app/sidecar/evaluate`
- **Command Center Dashboard**: `https://acn-fastapi-backend-322603900775.us-central1.run.app/dashboard`

## Developer Tooling & SDKs
- **Python**: `pip install btp-guard`
- **TypeScript / Node.js**: `npm install btp-guard`
- **Go**: `go get github.com/ivegotahunnitonit/bartholomew/pkg/btp`
- **Model Context Protocol (MCP) Proxy**: `python -m src.mcp_gateway --server-cmd <downstream-mcp>`
- **Claude Desktop & Cursor Config**: `python setup_claude_desktop_mcp.py`

## Benchmarks & Telemetry
- **Micro-Rollback Restoration**: 2.30 µs (P50), 3.42 µs (P99)
- **Secret Scrubbing Efficacy**: 100.0% interception across 12 credential entropy formats
- **AST Static Trait Analysis**: 1.38 µs evaluation
- **Technical Whitepaper**: `https://raw.githubusercontent.com/ivegotahunnitonit/bartholomew/main/paper_v2_4.md`