FusionAL

MCP is powerful. Enterprise MCP deployment is a different problem entirely.

Every server you connect loads its full tool manifest into Claude's context window before processing a single token of your request. At 84 tools across several servers, that's15,540 tokens consumed at session start— before your team has asked anything. At scale, that's cost, latency, and an audit surface nobody can explain to legal.

FusionAL is aself-hosted MCP governance gatewaythat sits between your AI clients and your tool servers. One endpoint. Central auth. Tool-level policy enforcement. Full audit trail. Deployable in a single Docker command — including on Windows, where MCP has 6 documented failure modes the official docs don't cover.

Built for teams that have already said yes to MCP and now need to run it safely:

- Token control— filter tool exposure per client so context bloat doesn't scale with your server count
- Audit-ready operations— every tool call logged with caller, input, latency, and response status
- Policy enforcement— control what tools run, for whom, and under what conditions
- Windows-hardened— 6 documented failure modes solved, none of which appear in official MCP docs
- AI-powered server generation— describe a tool in plain English, FusionAL builds and registers it
- Done-for-you option— we deploy, govern, and manage it for you

The question isn't whether to adopt MCP. It's whether your deployment will survive contact with your security team.

- Docker Desktop (installed and running)
- Claude Desktop
- Python 3.11+
- Git

git clone https://github.com/JRM-FusionAL/FusionAL.git cd FusionAL python -m venv venv source venv/bin/activate # macOS/Linux # venv\Scripts\activate # Windows pip install -r core/requirements.txt

cd core python -m uvicorn main:app --reload --port 8009

- Windows:C:\Users\YourName\AppData\Roaming\Claude\claude_desktop_config.json
- macOS:~/Library/Application Support/Claude/claude_desktop_config.json
- Linux:~/.config/Claude/claude_desktop_config.json

⚠️Windows users:Do NOT use%APPDATA%in the path — Claude Desktop won't expand it. Hardcode the full path. Example:C:\Users\YourName\AppData\Roaming\Claude\claude_desktop_config.json

{ "mcpServers": { "fusional-gateway": { "url": "http://localhost:8009/sse" } } }

Step 4: Restart Claude Desktop & Test

Fully quit and reopen Claude Desktop. Open a new chat and try:

"Generate a Python MCP server that fetches the current weather for a given city."

FusionAL will generate, register, and run it automatically.

Build and run the full gateway in Docker:

docker build -t fusional . docker run -d -p 8089:8009 --name fusional fusional

Gateway available athttp://localhost:8089.

Claude Desktop / Any MCP Client │ ▼ (SSE or Streamable HTTP) FusionAL Gateway (:8009) │ ┌─────┴──────┐ ▼ ▼ FastAPI MCP Transport (/mcp) REST API ├── execute_code ├── generate_and_execute └── generate_mcp_project │ ▼ Docker Sandbox Executor (network isolated, memory capped) │ ▼ MCP Server Registry (JSON) ├── business-intelligence-mcp (:8101) ├── api-integration-hub (:8102) └── content-automation-mcp (:8103)

cd ../mcp-consulting-kit docker compose up -d

- ✅ Network isolation (--network none)
- ✅ Memory limits (default 128 MB, configurable)
- ✅ Process limits (max 64 processes)
- ✅ Read-only filesystem (except/tmp)
- ✅ No privilege escalation
- ✅ Non-root user execution

Whenmcp-consulting-kit/showcase-servers/common/security.pyis present, the gateway enables:

- API key authentication on/execute,/register,/generate
- Rate limiting
- CORS configuration

FusionAL documents and solves 6 Windows MCP failure modes not covered in official docs:
-

FusionAL started with a NetworkChuck video.

The dice-roller example and quick-start structure in this repo are adapted from that tutorial (MIT licensed).

Chuck's tutorial shows you how to build MCP servers manually — edit a prompt template, paste it into Claude, implement the result yourself.

FusionAL removes the human from that loop entirely. Instead of a prompt file you copy-paste,generate_mcp_projectis a live tool call — Claude describes what it wants built, FusionAL generates it, registers it, and runs it. No clipboard. No manual steps.

More importantly, Chuck's video covers Docker's MCP catalog — 300+ servers you can enable with a click. That's step one. FusionAL is what you need when a CTO asks:

"Great — but how do we audit what those tools are doing? How do we enforce policy? How do we make sure this doesn't become a security incident?"

That's the gap FusionAL was built to fill: governance, auditability, and operational control for teams deploying MCP at scale — especially on Windows, where the catalog alone isn't enough.

Magica is your all-in-one AI platform, offering 2500+ cutting-edge tools under a single subscription.

Dynamically search and call tools using UnifAI Network

Connects AI agents to the SS&C Next Generation REST API to automate business processes.

Agentic pipeline that transforms ideas to revenue — for solo founders and bootstrappers.

The Most Comprehensive ServiceNow AI Toolkit — 400+ MCP tools, 26 AI capabilities, SDK mode, Direct API mode. Covers ITSM, ITOM, CMDB, HRSD, CSM, SecOps, GRC and 35+ modules.

200+ production tools for AI agents via Remote MCP. GitHub, Slack, Notion, Linear, Figma, Stripe, web search, OCR, document conversion, semantic memory/recall, Finviz, SEC filings, image generation, and more.

Create notes, search, & think with your Fabric AI workspace

Turn your Make scenarios into callable tools for AI assistants.

Connect to the Taskade platform via MCP. Access tasks, projects, workflows, and AI agents in real-time through a unified workspace and API.