openterms-mcp
Ed25519-signed consent receipts + programmable policy engine for AI agents before they take actions.
Openterms
Open source cryptographic consent receipts + programmable guardrails + provider verification for AI agents.
Your agent proves what it agreed to. Your policy controls what it's allowed to do. The API provider can verify both.
License Tests
Live Demo · Quickstart · Self-Host · MCP Server · Contributing · Open Receipt Spec
What It Does Openterms sits between your AI agent and the actions it takes. Three layers:
-
Receipts — Before your agent calls an API, it gets an Ed25519-signed receipt. Canonical JSON (RFC 8785), SHA-256 hash, real cryptography. Anyone can verify it using public keys — no API key needed, no trust in the server required.
-
Policy Engine — Daily spending caps, action type whitelists, escalation thresholds. The policy engine evaluates before the receipt is signed. Denied actions never get a receipt.
-
Provider Verification — API providers register their terms URL and verify agent consent before serving requests. One public GET call. Both sides of the transaction trust the proof.
Quickstart 60 seconds to your first receipt:
git clone https://github.com/jstibal/openterms.git cd openterms pip install flask pyjwt cryptography pyyaml bash quickstart.sh Or with Docker:
git clone https://github.com/jstibal/openterms.git cd openterms docker compose up --build
In another terminal:
bash quickstart.sh Receipt issuance is free — no wallet, no deposit, no payment required.
Self-Host Run your own Openterms instance:
Option 1: Direct
pip install flask pyjwt cryptography pyyaml python run.py
Server at http://localhost:5000
Option 2: Docker
docker compose up --build
Option 3: Point MCP server at your instance
export OPENTERMS_API_URL=http://localhost:5000 python openterms_mcp_server.py Everything runs locally. SQLite database, no external dependencies.
Hosted Service Don't want to self-host? Use the hosted instance at openterms.com — same open source code, managed for you.
MCP Server 10 tools for AI agents:
Tool What it does issue_receipt Signed receipt before any action, with provider verification headers verify_receipt Verify receipt cryptographic integrity (public) verify_receipt_by_hash Look up and verify by canonical hash (public) check_balance Workspace balance get_pricing Per-receipt pricing list_receipts Recent receipts get_policy Active guardrails — call on startup simulate_policy Pre-check: would this action be allowed? policy_decisions Audit trail of every allow/deny/escalate provider_activity Receipt stats for your API (provider auth) MCP Config { "mcpServers": { "openterms": { "command": "python3", "args": ["openterms_mcp_server.py"], "env": { "OPENTERMS_API_URL": "https://openterms.com", "OPENTERMS_API_KEY": "openterms_your_key_here" } } } }
How Provider Verification Works Agent Openterms API Provider | | | |-- issue_receipt ------->| | |<-- receipt + headers ---| | | |-- webhook notification -->| | | | |-- API call + headers ---|-------------------------->| | |<-- verify/{hash} ---------| | |-- receipt data ---------->| |<--------- response -----|---------------------------|
Agent issues receipt → gets X-Openterms-Receipt header Agent includes header in API call Provider calls GET /v1/receipts/verify/{hash} — public, no auth Valid → serve. Invalid → reject.
API Endpoints
Core Method Path Auth Description POST /v1/receipts Bearer Issue signed receipt POST /v1/receipts/verify None Verify receipt GET /v1/receipts/verify/{hash} None Verify by hash GET /v1/receipts Bearer List receipts GET /.well-known/jwks.json None Public signing keys Policy Engine Method Path Auth Description GET /v1/policy Bearer Active policy PUT /v1/policy Admin Create/update policy POST /v1/policy/simulate Bearer Test hypothetical action GET /v1/policy/decisions Bearer Decision audit trail Provider Verification Method Path Auth Description POST /v1/providers None Register as provider POST /v1/providers/verify Provider Verify domain GET /v1/provider/stats Provider Receipt stats GET /v1/provider/receipts Provider Recent receipts
Tests
make test
120 tests passing (80 core + 40 provider verification)
Architecture openterms/ ├── app.py # Flask API (1135 lines) ├── db.py # SQLite database (15 tables) ├── openterms_mcp_server.py # MCP server + CLI (10 tools) ├── core/ │ ├── canonical.py # RFC 8785 canonicalization │ └── signing.py # Ed25519 signing + JWKS ├── services/ │ ├── receipt_service.py # Receipt pipeline │ ├── ledger_service.py # Balance tracking │ └── policy_engine.py # Rule evaluation ├── tests/ │ ├── test_core.py # 80 tests │ └── test_mvp3.py # 40 tests ├── Dockerfile ├── docker-compose.yml ├── quickstart.sh └── .env.example
Roadmap
Phase Status What it does MVP1 ✅ Shipped Signed receipts — record what happened MVP2 ✅ Shipped Policy engine — enforce what's allowed MVP3 ✅ Shipped Provider verification — both sides trust the proof
ORS Spec 🔄 In progress Open Receipt Specification — portable format
Integrations 🔄 In progress LangChain, CrewAI one-line callbacks
MVP4 Planned Receipt chaining, agent certification
Contributing
See CONTRIBUTING.md. We especially welcome framework integrations, language SDKs, and feedback on the Open Receipt Specification.
License Apache 2.0 — see LICENSE.
Copyright 2026 Staticlabs Inc.
Related Servers
AgentRouter
Let your agent delegate tasks to specialised external agents and orchestrate multi agent approaches to tackle complex tasks and enable new capabilitys.
Nefesh Human State
Fuses biometric signals into a stress score (0-100) for real-time AI adaptation. MCP + A2A native.
TabNews Integration
Access data from the TabNews API.
Smithsonian Open Access
An MCP server to interact with the Smithsonian’s Open Access collection.
AgentBazaar
A2A marketplace with 1800+ AI agents. Search, invoke, trade capabilities, datasets, prompts.
FinMCP
Lightweight TypeScript Finance MCP server wrapping Yahoo Finance APIs. Plug real-time financial data — stocks, options, crypto, earnings — into any AI assistant. No API key. Works via stdio, Docker, or HTTP.
Memory Bank MCP
A production-ready Model Context Protocol (MCP) server that provides a powerful, vector-native memory bank for AI agents. Built with the Protocol-Lattice Go Agent Framework, this server offers persistent, searchable, and shareable memory with multiple database backends.
Control4 MCP Server
A safe-by-default MCP server that exposes your Control4 home automation (lights, scenes, locks, thermostats, and media) as structured tools over HTTP and Claude Desktop STDIO for reliable AI-powered control on your local network.
Librarian Nexus
Librarian Nexus is a shared knowledge library for AI agents. Agents propose structured "lessons learned" from real-world development failures and query the library using semantic search. Lessons include epic (what was attempted), failure (what went wrong), and solution (what fixed it). Queries are paywalled at $0.01 USDC via the x402 payment protocol on Base.
FastMCP Calculator Server
A calculator server that performs basic math operations like addition, subtraction, multiplication, division, power, and square root.