GiljoAI MCP

Context engineering platform for AI-assisted software development. Define your product once. Every agent that connects gets the full picture.

Free Community Edition · Self-Hosted · Privacy First · Bring Your Own AI

Works with: Claude Code CLI · Claude.ai / Desktop · Codex CLI · ChatGPT.com · Gemini CLI · Antigravity CLI · Any MCP Client

Website · Getting Started · Documentation · User Guide

GiljoAI MCP is a passive context server for AI coding tools. It stores your product knowledge, generates structured prompts, and coordinates multi-agent workflows via the Model Context Protocol (MCP). It does not write code or call any AI model. Your AI coding tool does all reasoning and coding using your own subscription.

GiljoAI sits at the intersection of product thinking and development. Whether you are a developer learning to define what you build before you build it, or a product manager turning a specification into working software, the platform gives you a structured path from vision to execution. The clearer your product definition, the more effective every agent session becomes.

Bring Your Own AI. GiljoAI works with Claude Code CLI, Claude.ai / Desktop, Codex CLI, ChatGPT.com, Gemini CLI, Antigravity CLI, or any MCP-compatible tool. Each connects with its own API key. You can run multiple tools simultaneously.

Windows (PowerShell):

irm giljo.ai/install.ps1 | iex

Linux / macOS:

curl -fsSL giljo.ai/install.sh | bash

The installer checks for prerequisites, downloads the latest release, sets up the database, builds the frontend, and walks you through configuration. When done, run python startup.py to start the server. First run opens the Setup Wizard in your browser.

git clone https://github.com/giljoai/GiljoAI_MCP.git
cd GiljoAI_MCP
python install.py
python startup.py

Requires Python 3.12+, PostgreSQL 18+, Node.js 22+.

Your Tools, Your Subscription

GiljoAI never touches your AI credits. You bring your own Claude Code CLI, Claude.ai / Desktop, Codex CLI, ChatGPT.com, Gemini CLI, Antigravity CLI, or any MCP-compatible tool, each with your own subscription. GiljoAI acts as a passive MCP server: your tool connects over HTTP, reads context and coordination data, and does all the reasoning and coding itself.

Define Your Product

Create a Product to represent the software you are building. Fill in context fields: description, tech stack, architecture, testing strategy, and more. Upload a vision document and let your AI tool populate the fields automatically. Vision documents are the most impactful input you can give GiljoAI. A well-written product proposal gives every agent session a shared understanding of what you are building and why.

Projects and Missions

1. Create a project and describe what needs to be done.

2. Activate the project. GiljoAI assembles a bootstrap prompt from your product context, 360 Memory, and project description.

3. Paste the prompt into your CLI tool. The orchestrator plans the mission and spawns agents.

4. Agents report status back in real time. You monitor progress on the Jobs page.

5. When the project completes, GiljoAI writes a 360 Memory entry. The next project inherits that accumulated context.

Skills and Agent Templates

One skill is installed on your machine during setup. Use it from your CLI without breaking flow:

Agent templates install and refresh through the giljo_setup tool (choose "Agents only"), not a slash command. The Agent Template Manager in the dashboard lets you customize agent profiles with roles, expertise, and chain strategies. Templates export in the correct format for your connected platform.

360 Memory

Each completed project writes to 360 Memory automatically: what was built, key decisions, patterns discovered, and what worked. This is not a plugin; it is a core product behavior. Your next project starts with accumulated context from previous ones. You control how many memories back agents read through the context settings. Optionally enrich memory with git commit history.

Dashboard and Monitoring

The Jobs page shows real-time agent activity: status, step progress, duration, and messages waiting. A message composer lets you talk directly to the orchestrator or broadcast to the entire agent team. All messages are logged in the MCP message system for auditability.

This is the GiljoAI MCP Community Edition, free for single-user use under the Elastic License 2.0.

Multi-user use requires a commercial license.

Production (single port):
  Browser --> :7272 (FastAPI) --> API + WebSocket + MCP + Static files
                               --> PostgreSQL (localhost:5432)

Development (two ports):
  Browser --> :7274 (Vite HMR) --> proxies /api, /ws, /mcp --> :7272 (FastAPI)
                                                             --> PostgreSQL (localhost:5432)

Database always runs on localhost. Auth is required for all connections. Multi-tenant isolation is enforced at the database level on every query.

python install.py              # Production install (recommended)
python install.py --dev        # Developer install (adds pre-commit hooks, NLTK data)
python install.py --headless   # Non-interactive mode for CI/CD

python startup.py              # Auto-detect mode and start
python startup.py --dev        # Development mode with Vite hot-reload
python startup.py --setup      # Re-run the Setup Wizard
python startup.py --no-browser # Start without opening the browser
python startup.py --verbose    # Detailed logging

API docs are served at runtime: Swagger UI at /docs, ReDoc at /redoc.

• JWT authentication required for all connections (no IP-based bypass)

• bcrypt password hashing (cost factor 12), minimum 12 characters

• 4-digit recovery PIN for self-service password reset (no email needed)

• Database always on localhost, never exposed to the network

• Plain HTTP by default; opt-in HTTPS (bring-your-own certificate) in Settings → Network

• Multi-tenant isolation on every database query

• Rate limiting on authentication endpoints

• Issues: github.com/giljoai/GiljoAI_MCP/issues

• Website: giljo.ai

• License: Elastic License 2.0

• Contributing: See CONTRIBUTING.md

GiljoAI MCP is built by GiljoAI.