kaginet
Kaginet - bitcoin escrow for AI agents
Documentation
Kaginet — Trustless Bitcoin Escrow for AI Agents
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
| Property | How |
|---|---|
| No custody | Mathematical lockout: the issuer key is destroyed before funding. Bitcoin consensus enforces this, not policy. |
| Private | Off-chain bearer transfers are invisible. No public ledger for intermediate transfers. |
| Hardware-attested | Intel TDX proves the code that deleted the key is the audited code. Verifiable against Intel's root CA. |
| 29 MCP tools | Any MCP-compatible agent (Claude, GPT-4, Cursor, Windsurf) gets payment capability with one URL. |
| Bitcoin mainnet | Real 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-langchainfrom 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 agentOption C: CrewAI
pip install kagikai-crewaifrom 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 keys1. 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 / RefundedFull state machine documentation: docs/instrument-lifecycle.md
Documentation
| Document | Description |
|---|---|
| Architecture | System components, trust boundaries, data flow |
| Instrument Lifecycle | State machine with all transitions |
| API Reference | Every REST endpoint with schemas |
| MCP Tools | All 29 tool definitions |
| Attestation | TDX attestation chain and verification |
| Dispute Mechanism | Optimistic settlement with bonds |
| Fee Model | Fee structure and estimation |
| Security Model | Threat model and trust boundaries |
| Comparison | Kaginet vs ERC-8183 vs Stripe vs AURA |
| Protocol Specification | Full escrow protocol: state machine, evaluator schemas, Nostr events, MCP tools |
Live Endpoints
| Endpoint | URL |
|---|---|
| MCP (SSE) | `https://mcp.kaginet.com/sse` |
| Developer Dashboard | cloud.kaginet.com |
| Product Website | kaginet.com |
Examples
- quickstart.py — Create your first escrow
- agent-escrow-flow.py — Full agent-to-agent payment
- mcp-config.json — MCP client configuration
- langchain-agent.py — LangChain agent with payment tools
- crewai-crew.py — CrewAI crew with escrow capability
- verify-attestation.py — Verify a TDX attestation
- fee-estimate.py — Check fees before creating an escrow
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 toolThe 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.
Frequently asked questions
What is kaginet?
kaginet is Kaginet - bitcoin escrow for AI agents
How do I install kaginet?
Open the GitHub repository and follow its README. Most MCP servers are added to your client's MCP config, then called by your agent.
Is kaginet open source?
Yes — it is hosted on GitHub at https://github.com/kaginetai/kaginet and has 1 stars.
Related MCP tools
cogneeCognee is the open-source AI memory platform for agents. Give your AI agents persistent long-term memory across sessions with a self-hosted knowledge graph engine.
skyvernAutomate browser based workflows with AI
hindsightHindsight: Agent Memory That Learns
watermarks-removerA privacy-first app that strips AI watermarks from content you own.
Qwen-AgentAgent framework and applications built upon Qwen>=3.0, featuring Function Calling, MCP, Code Interpreter, RAG, Chrome extension, etc.
pal-mcp-serverThe power of Claude Code / GeminiCLI / CodexCLI + [Gemini / OpenAI / OpenRouter / Azure / Grok / Ollama / Custom Model / All Of The Above] working as one.
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP