mcpField
K
Kaginet   Bitcoin Escrow For Ai Agents

Kaginet Bitcoin Escrow For Ai Agents

Remote
github.com/kaginetaiFinance1Updated 22w ago
Repository

About Kaginet Bitcoin Escrow For Ai Agents

Kaginet provides Bitcoin escrow tools for AI agents through MCP and framework adapters. Its workflow uses a hardware-attested TDX environment and Bitcoin mainnet escrow.

Key features

  1. 0129 payment and escrow tools
  2. 02Create, fund, and release escrow
  3. 03Verify TDX attestation
  4. 04LangChain and CrewAI adapters

Use cases

  1. 01Escrow payments between agents
  2. 02Hold payment for contracted work
  3. 03Check escrow status and release funds

MCP support

An MCP server exposes 29 payment and escrow tools to compatible clients. It is available at https://mcp.kaginet.com/sse.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://mcp.kaginet.com/sse
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

All endpoints
Streamable HTTPhttps://mcp.kaginet.com/sseAuth required

Tools

Details

From the project README

Kaginet — Trustless Bitcoin Escrow for AI Agents

CI PyPI - kagikai-langchain PyPI - kagikai-crewai License

Hardware-attested Bitcoin escrow that AI agents can use via MCP, LangChain, or CrewAI. No custody, no trust assumptions, no counterparty risk.

Live on Bitcoin mainnet. Running inside an Intel TDX confidential VM.


Why Kaginet

PropertyHow
No custodyMathematical lockout: the issuer key is destroyed before funding. Bitcoin consensus enforces this, not policy.
PrivateOff-chain bearer transfers are invisible. No public ledger for intermediate transfers.
Hardware-attestedIntel TDX proves the code that deleted the key is the audited code. Verifiable against Intel's root CA.
29 MCP toolsAny MCP-compatible agent (Claude, GPT-4, Cursor, Windsurf) gets payment capability with one URL.
Bitcoin mainnetReal money. Live now. Not a testnet demo.

Quick Start

Option A: MCP (recommended, zero install)

Add to your Claude Desktop, Cursor, or Windsurf MCP config:

{
  "mcpServers": {
    "kagikai": {
      "url": "https://mcp.kaginet.com/sse",
      "headers": {
        "Authorization": "Bearer kagi_YOUR_API_KEY"
      }
    }
  }
}

Get an API key at cloud.kaginet.com.

Your agent now has 29 payment tools. Try:

Create a 50,000 sat escrow to bc1q...recipient for "Logo design work"
Option B: LangChain
pip install kagikai-langchain
from kagikai_langchain import KagikaiToolkit

toolkit = KagikaiToolkit(
    base_url="https://mcp.kaginet.com",
    api_key="kagi_YOUR_API_KEY",
)
tools = toolkit.get_tools()  # 29 tools ready for your agent
Option C: CrewAI
pip install kagikai-crewai
from kagikai_crewai import (
    KagikaiEscrowCreateTool,
    KagikaiEscrowStatusTool,
    KagikaiEscrowReleaseTool,
)

# Add to your crew's agent
tools = [
    KagikaiEscrowCreateTool(),
    KagikaiEscrowStatusTool(),
    KagikaiEscrowReleaseTool(),
]

How It Works

Agent Runtime                MCP Server              ICS (TEE Enclave)           Bitcoin
─────────────                ──────────              ─────────────────           ───────
                                                                                    
  kagikai_escrow_create ──▶  Authenticate ──▶  Generate keys inside TDX            
                                               Destroy issuer key (kA)             
                                               Create P2TR escrow address          
                                               Generate bc1q funding addr          
                             ◀── instrument_id + funding_address ◀──               
                                                                                    
  "Fund this address" ────────────────────────────────────────────────▶ bc1q funded
                                                                                    
                                               Watcher detects funding             
                                               Auto-forward bc1q → bc1p    ──────▶ P2TR escrow
                                               Status: Confirmed                   
                                                                                    
  kagikai_escrow_release ─▶  Sign JWT ──────▶  Evaluate evidence                   
                                               Status: Completed                   
                                               Auto-sweep to payee         ──────▶ Payee receives
                                               Zero all keys                       
  1. Create: Agent calls kagikai_escrow_create with amount and payee address. ICS generates keys inside the TEE, destroys the issuer key, returns a bc1q funding address.
  2. Fund: Payer sends bitcoin to the bc1q address from any wallet. ICS auto-forwards to the internal P2TR escrow.
  3. Release: Agent submits evidence. Evaluator validates (auto or manual). ICS auto-sweeps funds to the payee.
  4. Attest: At any point, anyone can verify the TDX attestation proving the issuer key was deleted by the audited code.

Instrument Lifecycle

Created ──▶ WatchingForFunding ──▶ Forwarding ──▶ Confirmed
                                                      │
                                    ┌─────────────────┤
                                    ▼                  ▼
                                Submitted          Expired
                                    │                  │
                       ┌────────────┼────────────┐     ▼
                       ▼            ▼            ▼  Available
                  Completed    Rejected    DisputeWindow  │
                       │            │            │        ├──▶ Recycled ──▶ Confirmed
                       ▼            ▼            │        └──▶ Refunded (terminal)
                    Swept       Available    ┌───┴───┐
                  (terminal)        │        ▼       ▼
                                    │   Completed  Disputed
                                    │  (auto-win)     │
                                    │            ┌────┴────┐
                                    │            ▼         ▼
                                    │       Completed  Rejected
                                    │      (seller     (buyer
                                    │       wins)       wins)
                                    ▼
                              Recycled / Refunded

Full state machine documentation: docs/instrument-lifecycle.md


Documentation

DocumentDescription
ArchitectureSystem components, trust boundaries, data flow
Instrument LifecycleState machine with all transitions
API ReferenceEvery REST endpoint with schemas
MCP ToolsAll 29 tool definitions
AttestationTDX attestation chain and verification
Dispute MechanismOptimistic settlement with bonds
Fee ModelFee structure and estimation
Security ModelThreat model and trust boundaries
ComparisonKaginet vs ERC-8183 vs Stripe vs AURA
Protocol SpecificationFull escrow protocol: state machine, evaluator schemas, Nostr events, MCP tools

Live Endpoints

EndpointURL
MCP (SSE)https://mcp.kaginet.com/sse
Developer Dashboardcloud.kaginet.com
Product Websitekaginet.com

Examples


Project Structure

kaginet/
  adapters/
    langchain/          # pip install kagikai-langchain (29 tools)
    crewai/             # pip install kagikai-crewai (29 tools)
    tests/              # 90 adapter tests
  docs/                 # Architecture, API reference, protocol docs
  examples/             # Integration examples
  verify/               # TDX attestation verification tool

The ICS server (Rust, runs inside Intel TDX) is not included in this repository. This repo contains the integration layer: framework adapters, documentation, and examples.


Security

See SECURITY.md for our responsible disclosure policy.

The security model is documented at docs/security-model.md.


Contributing

See CONTRIBUTING.md. We welcome adapter contributions for new frameworks, documentation improvements, and example code.


License

See LICENSE. License terms are currently under evaluation.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.