Project Relay MCP Server
Coordinates multiple AI systems and human reviewers using Git as an inspectable record, providing tools for task management, evidence tracking, and independent review.
README
Project Relay
Project Relay is an evidence-governed coordination layer for multiple AI systems and human reviewers. It uses a Git repository as a shared, inspectable record while MCP provides a provider-neutral integration surface.
Status: pre-alpha M1 candidate. The name is a working title, the protocol is unstable, and no production deployment is implied.
Why it exists
Multi-model work usually fragments across chats, vendors, local tools, and copied prompts. Relay gives that work a common protocol:
- one explicit question per task;
- named owners, reviewers, and decision authorities;
- immutable evidence references and reproducible commands;
- independent review before acceptance;
- recorded disagreement instead of silent consensus;
- human approval for consequential decisions;
- no requirement for every model to share one vendor API.
Relay coordinates work. It does not decide whether a scientific claim is true, replace peer review, or make model output trustworthy by default.
Architecture at a glance
AI clients / humans / IDE agents
|
MCP interface
|
Relay protocol + policy
|
append-only events and evidence
|
Git repository
|
read-only web console / GitHub UI
The canonical record is validated JSON plus referenced artifacts in Git history. Issues, labels, dashboards, and the Pages console are projections of that record, never alternate sources of truth.
Current public milestone
The local M1 vertical slice now includes:
- JSON Schemas for tasks, events, evidence, reviews, and decisions;
- deterministic canonicalisation and SHA-256 provenance helpers;
- append-only event-chain verification plus policy-checked task transitions;
- an evidence-bundle CLI that hashes declared artifacts without executing commands;
- a read-only MCP server with tools, a task resource, and an independent-review prompt;
- a deterministic static console projection of state, history, and gate results;
- a synthetic submission, review, remediation, resubmission, and human-decision fixture;
- Windows and Linux CI conformance coverage plus a public-boundary scanner.
GitHub write automation, remote MCP transport, authentication, and hosted multi-tenant operation are deliberately deferred until their threat models are reviewed.
Quick start
Requires Node.js 24 or newer.
npm install
npm run check
npm run mcp
Check direct dependency versions without changing the repository:
npm run versions:check
Run the offline local doctor for measured runtime, Git, lockfile, workspace, and worktree checks:
npm run doctor
To review conservative updates and receive an interactive [y/N] installation prompt:
npm run versions:update
The MCP process uses stdio and reads the workspace specified by RELAY_WORKSPACE (the current directory by default):
{
"mcpServers": {
"project-relay": {
"command": "npm",
"args": ["--prefix", "/absolute/path/to/Project-Relay", "run", "mcp"],
"env": {
"RELAY_WORKSPACE": "/absolute/path/to/a/relay-workspace"
}
}
}
}
Use examples/m1 to inspect the complete local remediation loop. examples/minimal remains the smallest M0 kernel fixture.
Create a validated evidence bundle from an explicit manifest:
npm run evidence:create -- --manifest path/to/manifest.json --output path/to/bundle.json
The command reads and hashes declared artifacts. It does not execute the commands recorded in the manifest.
Safety and publication boundary
This public repository must never contain credentials, private datasets, unpublished ITSM material, personal information, confidential prompts, or commercial records. .gitignore is not a security boundary: releases must be exported from a separate private workspace using an explicit allowlist and reviewed as if every commit were permanent.
See Publication Boundary and Security before contributing.
Documentation
- Architecture
- Protocol and decision gates
- Public/private publication boundary
- Version maintenance
- Roadmap
- Contributing
- Licence status
Licence status
No open-source licence has been selected yet. Publication on GitHub does not grant permission to reuse the work beyond rights provided by GitHub's terms. See LICENSE-STATUS.md.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.