MCP Servers

模型上下文协议服务器、框架、SDK 和模板的综合目录。

The first open-source, agent-first financial ledger built for autonomous AI agents using .NET 8, Python, and the Model Context Protocol (MCP).

创建于 5/12/2026
更新于 about 5 hours ago
Repository documentation and setup instructions

Overview AgentVault is a production-grade treasury management system designed specifically for the Agentic Economy. Unlike traditional banking APIs built for human-in-the-loop interfaces, AgentVault provides a secure, machine-readable "Vault" that allows autonomous agents to manage financial operations within strict, verifiable human policies.

This project bridges the gap between high-level AI reasoning and low-level financial integrity.

Key Features 🏦 Immutable Double-Entry Ledger: A senior-architected .NET 8 backend ensuring every transaction is mathematically balanced and tamper-proof.

🧠 MCP Native (Model Context Protocol): Built-in support for the Anthropic/Claude MCP standard, allowing agents to "see" and "interact" with the bank natively via stdio.

🛡️ Policy-Driven Guardrails: A circuit-breaker middleware that intercepts agent requests to enforce spending limits, vendor whitelisting, and mandatory reasoning audits.

📜 Reasoning Audit Traces: Every transaction requires a reasoning_trace. This creates a dataset of "Intent vs. Action," critical for fraud detection and AI safety research.

🚦 Human-in-the-Loop (HITL): Seamless escalation logic where high-value or high-risk agent proposals are paused for a manual human signature.

The Tech Stack Backend: .NET 8 Web API (C#)

Agent: Python 3.11+ (using the mcp and tenacity SDKs)

Database: PostgreSQL (Double-entry schema)

Protocol: Model Context Protocol (MCP) over stdio

Why This Exists As we transition from "Chatbots" to "Autonomous Agents," the bottleneck is Trust. Companies cannot give an LLM a credit card. AgentVault solves this by providing a programmatic "Allowance" and a strictly governed environment where the system, not the LLM, is the ultimate authority over the money.

Roadmap & Academic Research This project is currently being expanded as part of a Master’s in Data Science at the Norwegian University of Life Sciences (NMBU). Future updates will include:

Predictive Fraud Detection: Using reasoning logs to identify "Agent Hallucination" in financial proposals.

Cross-Border Settlement: Integrating Norwegian Open Banking APIs (DNB/Nordea) for real-world Nordic settlements.

License Licensed under the Apache License 2.0. See the LICENSE file for details.

快速设置
此服务器的安装指南

安装命令 (包未发布)

git clone https://github.com/110011193/Treasury-MCP
手动安装: 请查看 README 获取详细的设置说明和所需的其他依赖项。

Cursor 配置 (mcp.json)

{ "mcpServers": { "110011193-treasury-mcp": { "command": "git", "args": [ "clone", "https://github.com/110011193/Treasury-MCP" ] } } }