MCPCloud

Self-hosted MCP (Model Context Protocol) gateway. Write any Python function, register it as a skill, and it instantly becomes a tool that Claude Desktop, Claude API, Cursor, or any MCP-compatible client can call.

Website: mcpcloud.dev · GitHub: carsor007/mcpcloud

How it works

Claude Desktop / Claude API / any MCP client
            ↓  MCP tool call
        MCPCloud  (this repo)
            ↓  your code runs
    Jira · Slack · Salesforce · anything

Every Python function registered as a skill becomes an MCP tool. No vendor lock-in, no proprietary agent framework — just functions.

Related MCP server: Xiaobai Print MCP

Quickstart

Run locally with Docker:

git clone https://github.com/carsor007/mcpcloud.git
cd mcpcloud
cp .env.example .env
docker compose up

Open http://localhost:8000/ui — the tool browser shows all registered skills.

Connect Claude Desktop:

curl http://localhost:8000/mcp/jira_ops/config

Paste the returned JSON into ~/Library/Application Support/Claude/claude_desktop_config.json under "mcpServers". Restart Claude Desktop.

Adding a skill

Drop a .py file into skills/. Any file with a register_all() function is loaded automatically on startup.

# skills/my_tools.py
from registry import SkillResult, get_registry

async def my_skill(input: dict, context: dict) -> SkillResult:
    '''One-line description shown in the UI.'''
    return SkillResult(success=True, output={"result": input.get("text", "")})

def register_all():
    get_registry().register(
        "my_tools",   # agent type  — groups skills in the sidebar
        "my_skill",   # skill name  — shown under the group
        my_skill,
        schema={
            "type": "object",
            "required": ["text"],
            "properties": {
                "text": {"type": "string", "description": "Input text"}
            }
        }
    )

Restart the server. The skill appears in the UI and is immediately callable as an MCP tool.

Included skills

Both work out of the box — real API calls run when credentials are configured, stub data is returned otherwise.

jira_ops

Configure by setting JIRA_URL, JIRA_EMAIL, JIRA_API_TOKEN in .env.

slack_ops

Configure by setting SLACK_WEBHOOK_URL in .env.

Deploy to AWS

The Deploy to AWS button above launches a CloudFormation stack in your account. The public image is already hosted — no ECR setup required.

It provisions:

• ECS Fargate — 2 tasks, no EC2 to manage

• ElastiCache Redis — session tracking across workers (~$15/month for t4g.micro)

• Application Load Balancer — with optional HTTPS via ACM

• Secrets Manager — stores Jira, Slack, and API credentials securely

You will need: a VPC with at least 2 public subnets.

One-click deploy (console):

Click the Deploy to AWS button above.

CLI deploy:

cd deploy/aws
make deploy \
  VPC_ID=vpc-xxxxxxxxxxxxxxxxx \
  PUBLIC_SUBNETS=subnet-aaa,subnet-bbb \
  PRIVATE_SUBNETS=subnet-ccc,subnet-ddd

After deployment, retrieve your URLs:

make outputs

Self-hosting your own image (optional — if you've modified the code):

cd deploy/aws
make ecr-public-create        # one-time: creates public ECR repo
ECR_PUBLIC_ALIAS=your_alias make push-public

Configuration

License

Apache 2.0 — see LICENSE.

Free to self-host. Managed deployment available on AWS Marketplace.