Templates and install scripts for wiring up any MCP-compatible AI client to a useful set of Model Context Protocol servers — so your assistant can talk to AWS, Jenkins, Bitbucket, Atlassian, and your databases directly instead of through copy-paste.

Engineers who want their AI coding assistant to actually do work against real systems — read pod logs, look up a Jira ticket, inspect a queue's backlog, check a DB schema — not just chat about code. Wiring each MCP server up by hand is fiddly: every server has its own auth, env vars, and gotchas, and config snippets you find online are usually pinned to one specific AI client. This kit ships an opinionated, client-agnostic bundle of the MCP servers most teams need, plus reference docs written for the AI to read so it knows when to reach for which tool.

After setup (~15 min):

$ claude mcp list
Checking MCP server health…

aws-cloudwatch: uvx awslabs.cloudwatch-mcp-server@latest         - ✓ Connected
atlassian:      npx -y mcp-remote https://mcp.atlassian.com/v1/mcp - ✓ Connected
jenkins:        uvx mcp-jenkins                                  - ✓ Connected
aws-sqs:        uvx awslabs.amazon-sns-sqs-mcp-server@latest     - ✓ Connected
aws-athena:     uvx awslabs.aws-dataprocessing-mcp-server@latest - ✓ Connected
aws-rds:        uvx awslabs.aws-api-mcp-server@latest            - ✓ Connected
bitbucket:      uvx atlassian-bitbucket-mcp@latest               - ✓ Connected
mysql-db:       mysql-mcp-server                                 - ✓ Connected
mssql-db:       mcp-sqlserver                                    - ✓ Connected

(Output shown for Claude Code; equivalent status views exist in Cursor, Continue.dev, Cline, etc. — see SETUP.md § "Verify".)

Works with:

• Claude Code
• Cursor
• Continue.dev
• Cline
• Windsurf
• Zed (via its MCP support)
• VS Code MCP extensions
• Any custom app built with the official MCP SDKs

The MCP server config block ({"mcpServers": {...}}) is a cross-client standard; the per-tool wrapping differs only in where you paste it (see SETUP.md for the per-client mapping).

A base layer everyone needs, plus independent database modules so you only install what you use.

Different npm packages, different env-var conventions — shipped as independent modules so you install zero, one, or both.

mcp-starter-kit/
├── README.md                     ← you are here
├── SETUP.md                      ← step-by-step base setup + per-client wiring
├── LICENSE                       ← MIT
├── install.sh                    ← installs base layer (uvx, AWS CLI, MCP packages)
├── AGENTS.md.template            ← drop-in instructions for your AI assistant —
│                                   copy to your tool's instruction-file location
│                                   (CLAUDE.md, .cursorrules, AGENTS.md, etc.)
├── templates/
│   ├── aws-config.template       ← AWS SSO config skeleton
│   └── mcp-servers.json          ← cross-client MCP config block (no DB)
├── databases/
│   ├── mysql/   → README.md + install.sh + mcp-block.json
│   └── mssql/   → README.md + install.sh + mcp-block.json
└── docs/                         ← per-MCP reference, written FOR your AI to read
    ├── jenkins.md
    ├── bitbucket.md
    ├── atlassian.md
    ├── aws-cloudwatch.md
    ├── aws-athena.md
    ├── aws-sqs.md
    └── aws-rds.md

When you invite an AI assistant into your workflow, the limiting factor isn't "can the tool reach the system" — it's "does the assistant know which tool to reach for, and how." The docs/ directory is plain markdown your assistant can read; combined with AGENTS.md.template, it ends up knowing things like:

• "For application logs, use the CloudWatch MCP — don't shell out to aws logs."
• "If an AWS call returns Token has expired, run aws sso login and retry once before asking the user."
• "The SQS MCP defaults to us-east-1 — pass an explicit region= if your queues live elsewhere."

These aren't novel insights — they're the kind of operational knowledge a human teammate accumulates after a week. Front-loading it for your AI on day one is what an AGENTS.md / CLAUDE.md / .cursorrules / .clinerules file is for.

# 1. Install base
chmod +x install.sh && ./install.sh

# 2. Configure (see SETUP.md for details)
cp templates/aws-config.template ~/.aws/config         # then edit
# Edit templates/mcp-servers.json — fill in placeholders
# Wire it into your AI client (per-client steps in SETUP.md)

# 3. Add a database (only if you need one)
bash databases/mysql/install.sh   # for MySQL
bash databases/mssql/install.sh   # for MS SQL Server

# 4. Verify
#    Claude Code:   claude mcp list
#    Cursor:        Settings → Features → MCP Servers (UI shows status)
#    others:        check your client's MCP server / tool list

Full walkthrough: SETUP.md.

MIT — see LICENSE. Use it however you like.

PRs welcome. Especially:

• Additional MCP modules (Postgres, Redis, GitHub, GitLab, Kubernetes, …)
• Per-client wiring examples (more clients, Windows install scripts)
• Per-MCP doc improvements