MCP server directory
MCP servers I have built, used, or recommend, with honest notes on what each one is actually good for.
The Model Context Protocol is how AI agents talk to real systems. An MCP server is the adapter layer: it exposes your tools, APIs, and data as typed functions a model can call. This page is a working reference, not an aggregator list. Every entry here has been used in a real project or evaluated hands-on.
I am Ali Kazim, an AI engineer based in Melbourne who builds MCP servers for production use. If something on this list is wrong or outdated, that is a bug. If you want a server built for your system, see the MCP server development service.
Estimation tools
Two calculators for scoping AI and product work. No email required.
Fill out the form to get an estimate
Servers I have built (open source)
Use when: You want Claude to answer questions about your Swell store data without building a custom integration.
Do not use when: You need write operations (orders, fulfilment) in production without additional guardrails.
Use when: You need an AI assistant to operate across a closed ERP system with real inventory state.
Do not use when: You want a drop-in solution — this was purpose-built for a specific system and is not generic.
Servers worth knowing about (curated third-party)
A short list from the broader MCP ecosystem that I have evaluated and consider worth using. Quality over quantity. This list is updated periodically and is not exhaustive.
File system access
Local file system
Reference implementation from Anthropic. Good for understanding the protocol. Not a production security boundary — treat it as a local dev tool, not a server you expose.
Web search
Brave Search API
Solid for giving a model real-time web access. Requires a Brave API key. Works reliably in agentic loops where freshness matters.
Web search
Tavily Search API
AI-optimised search results with cleaner structured output than Brave. Preferred when you need the model to synthesise web results rather than just retrieve URLs.
Database
PostgreSQL database
Read-only Postgres queries via MCP. Functional for analytics and lookup use cases. Add your own auth layer before exposing to any production database.
Database
SQLite database
Lightweight local DB access. Good for prototype and offline agent workflows. Not appropriate for multi-user or server-side deployments.
Browser automation
Browser via Playwright
Microsoft-maintained browser automation MCP. Production-grade compared to most browser servers. Use for scraping, form filling, and UI-driven workflows.
MCP server examples by use case
Practical decision guide. Pick the scenario closest to your problem.
A founder wants their AI assistant to answer questions about business data — orders, customers, inventory — without manually exporting CSVs.
A read-only SQL tool server over stdio, pointed at your Postgres or SQLite database. The model sends queries, the server validates and executes them, and returns structured results. Start with the postgres server above as a reference, then add your own auth layer.
A team running fulfilment manually wants an agent that can look up current stock levels and trigger purchase orders through an ERP or inventory system.
A custom MCP server purpose-built for your ERP's API surface. Read tools for inventory queries, write tools for order creation behind a confirmation step. This is not a generic server you can download — it requires a build scoped to your system.
Requires a custom build. See the MCP server service for scope and pricing.
An agent needs up-to-date information — competitor pricing, news, public data — as part of a multi-step task.
Brave or Tavily search server depending on whether you need raw results or pre-synthesised content. Both are listed above. Drop one into your agent loop and pass search as a named tool.
A product team wants Claude to answer support questions, pull usage metrics, or summarise customer state using live application data.
A read-optimised MCP server that wraps your internal API with typed tool definitions. The key design decision is which tools to expose and what each one returns — over-fetching or under-specifying here is where most implementations fail.
Requires a custom build. See the MCP server service for scope and pricing.
Custom builds
Most production use cases require a server purpose-built for your systems and access patterns. I build them end to end, from tool schema design through deployment.