MCP Server. Model Context Protocol
MCP is now a Linux Foundation standard with 97 million monthly SDK downloads. Every Claude, every agent using MCP can connect to OMEGA governance in minutes. Every decision your agent makes gets a governed record: before it executes.
Current v1.4 (21 April 2026); prior v1.3 (20 April 2026).
The problem: Your AI agent makes decisions. Those decisions have consequences. Right now there is no governed record of what the agent was authorised to do, how it reasoned, or whether a gate existed before it acted.
What OMEGA MCP adds: Before your agent executes any consequential action, OMEGA produces a governed record: sealed, hash-chained, permanent. The gate fires. The record exists. The action is authorised or held.
v1.4 records include the 21-conjunct formal bundle verification flag and carry the four structural integrity constraints (P1 freshness, P2 DAG, P6 atomic agency, P4T environment invariant) as validation requirements.
Submit a governed decision record before any consequential action. Returns a SHA-256 sealed record with mandate ID.
POST /decisions
Retrieve any governed record by mandate ID. Immutable. Hash-verifiable. Survives removal of the system that produced it.
GET /decisions/:mandate_id
Run a governance gap analysis against any automation description. Returns missing primitives and regulatory exposure.
GET /diagnostic?desc=
Validate a proposed decision record against the fifteen OMEGA primitives. Returns PASS or the specific primitive that fails.
POST /validate
Add the OMEGA server to your Claude Desktop, Cursor, or any MCP-compatible client configuration file.
{
"mcpServers": {
"omega-governance": {
"command": "npx",
"args": [
"-y",
"@omega-protocol/mcp-server"
],
"env": {
"OMEGA_API_URL": "https://omega-governance-api-production.up.railway.app"
}
}
}
}
Every consequential decision your agent makes is submitted to OMEGA before execution. The governed record is returned. The action proceeds or is held. The record is permanent.
First external integration: Deliberate Network (Alex Stojanovic), SHA-256 hashed OMEGA record returned first attempt, March 2026. The standard works in production.
# Before executing: agent calls omega_submit_decision { "tool": "omega_submit_decision", "input": { "mandate_id": "VLRW6HUK", "governance": { "authorised_agent": "loan-approval-agent-v2", "authority_document": "LendingPolicy-v3.2", "constraint_envelope": "max_loan: 50000, min_score: 620" }, "reasoning": { "facts": ["credit_score: 687", "income_verified: 42000"], "inference": "applicant meets all policy criteria", "assumptions": ["income documents are accurate"], "unknowns": ["future employment stability"] }, "expectation": { "predicted_outcome": "loan repaid within 48 months", "confidence": "MEDIUM-HIGH", "falsifiable": true }, "confirmation": { "gate_type": "AUTOMATED", "irreversibility_acknowledged": true } } } # OMEGA returns immediately: { "status": "COMMITTED", "mandate_id": "VLRW6HUK", "record_hash": "sha256:a3f9e2c1d8b7...", "timestamp": "2026-04-19T14:23:07.002Z", "primitives_verified": 15, "protocol_version": "v1.4", "conjunct_count": 21 }
OMEGA publishes an Agent Card at the standard well-known location. Every A2A-connected agent can discover OMEGA automatically.
https://www.omegaprotocol.org/.well-known/agent.json
The agent card declares OMEGA's capabilities, fifteen primitives, honest limits, API endpoints, and regulatory coverage. Any agent that reads agent cards will find OMEGA and know what it governs.
MCP was donated to the Linux Foundation Agentic AI Foundation in late 2025. 97 million monthly SDK downloads by February 2026. Near-total consensus across OpenAI, Google, Microsoft, Amazon. It is the coordination layer for the agentic economy.
The problem it does not solve: every agent that connects via MCP can now take actions: but no standard governs whether those actions were authorised, reasoned, and gated before execution. OMEGA is the governance layer that MCP does not provide.
Trust properties that hold at one agent degrade at ten and collapse at a thousand: unless governance is native to every decision, not bolted on at the system level. OMEGA embeds fifteen primitives at the decision layer. Scaling the agent count scales the governance automatically.