p-e relay MCP server
Commit Sep 18, 2026Provides a multi-agent synchronization and dispute resolution protocol for autonomous LLM swarms using atomic append ledgers and adversarial adjudication.
Also known as it-start/relay-ui, relay-ui
- Stars
- 0
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- 0.1.0
- Last commit
- Sep 18, 2026
- License
- MIT
What is p-e relay MCP?
p-e relay provides a multi-agent synchronization and dispute resolution protocol for autonomous LLM swarms using atomic append ledgers and adversarial adjudication.
Summary generated from the project's README
About p-e relay MCP server
Synchronizing autonomous LLM swarms through a leaderless protocol is what p-e relay enables for developers building distributed agent systems. It provides a framework for Claude, GPT-4o, Gemini, and Mistral agents to communicate using atomic append ledgers and adversarial adjudication. The system ensures causal ordering across asynchronous nodes by using Hybrid Logical Clocks and resolves write races at the kernel level via POSIX atomic slot allocation.
Users interact with the protocol through a remote setup that includes an Interactive Workbench and a Live Relay Ledger. To enable autonomous reasoning and adjudication, the ALLOW_SERVER_MODEL_CALLS environment variable must be set to 1, allowing the server to use its own API keys for routes like /api/relay/adjudicate and /api/relay/agent-exec. The server supports real-time updates through Server-Sent Events and maintains data integrity using RFC 8785 canonicalization for cryptographic digests.
What you can do with it
- 1Resolve agent communication conflictsUse the adversarial adjudication protocol to settle disputes between competing LLM agents.
- 2Maintain strict message orderingEnsure all agents in a swarm follow a consistent causal sequence regardless of network drift.
- 3Prevent data collisions in ledgersHandle concurrent write attempts from multiple agents using atomic filesystem locks.
- 4Verify message integrity across platformsValidate that JSON payloads remain bit-exact across different programming languages and toolchains.
Written from the project's README. Every use case is checked against a quote from it.
How to install p-e relay MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall p-e relay MCP in Claude Code
$ claude mcp add --transport http relay-ui https://relay.zae.life/api/mcpInstall p-e relay MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"relay-ui": {
"url": "https://relay.zae.life/api/mcp"
}
}
}Install p-e relay MCP in VS Code
.vscode/mcp.json
{
"servers": {
"relay-ui": {
"type": "http",
"url": "https://relay.zae.life/api/mcp"
}
}
}Install p-e relay MCP in VS Code CLI
$ code --add-mcp '{"name":"relay-ui","type":"http","url":"https://relay.zae.life/api/mcp"}'Install p-e relay MCP in Codex
$ codex mcp add relay-ui --url https://relay.zae.life/api/mcpInstall p-e relay MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.relay-ui] url = "https://relay.zae.life/api/mcp"
Install p-e relay MCP in Gemini CLI
$ gemini mcp add -s user --transport http relay-ui https://relay.zae.life/api/mcpInstall p-e relay MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"relay-ui": {
"url": "https://relay.zae.life/api/mcp",
"type": "http"
}
}
}Automated checks
Ran Nov 15, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachableit-start/relay-ui is public and reachablegithub.com/it-start/relay-ui ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-09-18github.com/it-start/relay-ui ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeinitialize answered with protocol 2025-06-18 in 476 msPass
README
View on GitHubAgent Relay & Adjudication Protocol (v1.0-SPEC)
An industrial-grade, leaderless, multi-agent synchronization and dispute resolution protocol designed for autonomous LLM swarms (Claude, GPT-4o, Gemini, Mistral, and Human Architects).
Built upon canonical cryptographic digests (RFC 8785 / Proverbs 11:1), causal monotonicity (Hybrid Logical Clocks / HLC), atomic lock-free append ledgers (POSIX O_CREAT | O_EXCL), and adversarial cross-examination (Proverbs 18:17 Adjudication Court).
🏛️ Core Architectural Invariants (SPEC MUST 1–8)
- RFC 8785 JCS Canonicalization (MUST 1): Cryptographic SHA-256 digests are computed strictly on recursively sorted JSON keys, IEEE 754-normalized floats, and UTF-8 encoded payloads ("A false balance is an abomination to the Lord, but a just weight is his delight" — Proverbs 11:1).
- Hybrid Logical Clocks (MUST 2): Every envelope carries physical UTC time combined with a monotonic Lamport counter (
physical_ms:counter:node_id) guaranteeing causal ordering across asynchronous nodes without NTP drift failure. - Atomic Slot Allocation via
O_CREAT | O_EXCL(MUST 3): Concurrent write races are resolved at the kernel filesystem level. If two agents attempt to claim the same sequence slotseq: N, the collision returnsEEXIST, forcing the loser to incrementseqmonotonically. - Adversarial Triad & Adjudication (MUST 4): No claim is ratified without adversarial cross-examination ("The one who states his case first seems right, until the other comes and examines him" — Proverbs 18:17).
- Zero Implicit Trust (MUST 5): Inter-agent envelopes must be verifiable offline and independently validated across different toolchains (TypeScript, Python, Rust, Go).
- Known-Missing Retention (MUST 6): When an envelope payload is purged/compacted or lost, the ledger retains its sequence slot, locator, and SHA-256 digest with status
MISSINGto prevent phantom slot re-use. - Cross-Platform Rosetta Compatibility (MUST 7): Exact wire compatibility across JSON, YAML, Protocol Buffers, FlatBuffers, and MCP (Model Context Protocol).
- Real-time SSE & File Bus Streaming (MUST 8): Sub-millisecond reactive subscriptions via Server-Sent Events (SSE) and POSIX file watchers.
🚀 Quick Start
Prerequisites
- Node.js 20+ (the scripts run under it; this is what the hosting container has)
- Bun 1.1+ optional — a faster installer and runner for the same scripts
Installation & Run
# Clone the repository
git clone https://github.com/your-username/agent-relay.git
cd agent-relay
# Install dependencies — bun, because bun.lock is the lockfile here
bun install --frozen-lockfile
# Start development server (Port 3000)
bun run dev
# Build production bundle
bun run build
bun run start
Open http://localhost:3000 to access the full Interactive Workbench, Live Relay Ledger, and Autonomous Agent Chat.
The commands above are bun, and the scripts they run stay Node-compatible
on purpose. This is an AI Studio applet (metadata.json), and the hosting
container invokes npm run build and npm start in a Node image with no
bun on PATH. So npm is not a habit to be migrated away from here — it is
what one real deployment target does, and package.json requires no binary
either runtime lacks. CI keeps a node job for that reason and a bun job
because O_CREAT | O_EXCL is implemented by each runtime for itself.
Everything this repository controls — deploy.sh, the contributor steps, the
commands above — uses bun, and bun install --frozen-lockfile is the
deterministic install: bun.lock is the only lockfile committed, and npm would
write a second one that can disagree with it.
Environment
HOST |
Interface to bind. Defaults to 127.0.0.1. This process authenticates nothing, so anything wider is a deliberate act and belongs behind a reverse proxy that does. |
PORT |
Defaults to 3000. |
NODE_ENV |
production serves the built client from dist/. Anything else starts a Vite dev server, which is not meant to face a network. bun run start sets it. |
PE_STORE_ROOT |
Read an existing p-e relay store instead of this project's own. Read-only: write, delete and reset are declared unavailable and refused with 405. |
ALLOW_SERVER_MODEL_CALLS |
1 enables the three routes that run models on API keys held by this process — /api/relay/adjudicate, /api/relay/step-triad, /api/relay/agent-exec — none of which has authentication of its own. Off by default. ALLOW_AGENT_EXEC is accepted as the former name. |
📦 Project Structure
.
├── src/
│ ├── components/
│ │ ├── AgentChatInterface.tsx # Multi-agent chat with 1-click Triad & Court
│ │ ├── LiveRelayConsole.tsx # Real-time ledger, sequencer & SSE observer
│ │ ├── AdjudicationWorkbench.tsx # Court state machine (Proverbs 18:17)
│ │ ├── RosettaMatrix.tsx # Cross-language wire serializers
│ │ ├── EnvelopeStudio.tsx # RFC 8785 canonizer & digest generator
│ │ ├── FailureSandbox.tsx # Chaos injection & invariant stress tests
│ │ ├── BridgeExporter.tsx # MCP & Tool definitions export
│ │ └── Navbar.tsx # Responsive top navigation
│ ├── App.tsx # Root React 19 application
│ ├── main.tsx # Vite React mount point
│ └── index.css # Tailwind CSS & theme tokens
├── server.ts # Express + Atomic POSIX Store + SSE Hub
├── docs/ # Architecture specs & Engineering Backlog
│ ├── ARCHITECTURE.md # In-depth architectural blueprint
│ └── BACKLOG_AND_DEBT.md # Engineering insights, debt & roadmap
├── metadata.json # AI Studio runtime metadata
├── AGENTS.md # Context & rules for autonomous AI coding agents
├── CONTRIBUTING.md # Contribution guidelines
├── LICENSE # MIT License
└── package.json # Build scripts & dependency declarations
🛠️ API & Wire Protocol
Envelope Specification
{
"seq": 42,
"locator": "agent:claude/42@2026-09-01T16:45:00.000Z",
"from": "agent:claude-code",
"to": "agent:chatgpt-adversary",
"type": "claim",
"title": "HLC Monotonic Invariant Proposal",
"parent_locator": "agent:human/41@2026-09-01T16:44:00.000Z",
"hlc": "1788281100000:0001:claude",
"digest": "sha256:e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
"payload": {
"body": "Proposed RFC 8785 canonical digest validator."
}
}
Core HTTP Endpoints
POST /api/relay/deposit— Atomically deposit an envelope usingO_CREAT | O_EXCLGET /api/relay/records— Fetch current sequence ledger & envelope stateGET /api/relay/events— Real-time Server-Sent Events (SSE) feedPOST /api/relay/adjudicate— Execute Proverbs 18:17 Adjudication verdict via GeminiPOST /api/relay/agent-exec— Autonomous agent reasoning & response dispatcher
📜 Biblical & Mathematical Foundations
- Proverbs 11:1: "Dishonest scales are an abomination to the LORD, but a just weight is His delight." — Foundational rule for bit-exact JSON Canonicalization (RFC 8785).
- Proverbs 18:17: "The first to state his case seems right, until another comes and examines him." — Protocol mandate for adversarial cross-examination prior to ledger ratification.
📄 License
MIT License. Free for open-source research and enterprise distributed agent deployments.
p-e relay MCP questions
How do I add p-e relay MCP to Claude Code?
Run claude mcp add --transport http relay-ui https://relay.zae.life/api/mcp in your terminal. Add --scope user to make it available in every project.
How do I add p-e relay MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
How does the protocol handle concurrent write races?
Concurrent write races are resolved at the kernel filesystem level using atomic slot allocation via O_CREAT | O_EXCL.
Can I use an existing p-e relay store?
Yes, you can read an existing p-e relay store by setting the PE_STORE_ROOT environment variable, though writes and deletes will be refused.