A roundup of useful MCP servers for a desktop agent — GitHub, databases, Telegram, trackers, and calendar. What each one gets you and how to add a server in Doka in a couple of minutes.

MCP gives an agent hands: it's how a model connects to external services and starts working with them. But which servers are actually worth connecting? Here's a roundup of the ones that genuinely come in handy, with a plain explanation of why each one matters.

The MCP ecosystem is already large: ready-made servers exist for dozens of services, and almost all of them are open source. This isn't the whole list, just a sensible place to start.

GitHub

The agent gets direct access to repositories: reads issues, looks at pull requests, goes through commit history. Handy for "look at what changed on this branch" or "pull together the open tasks for this project" — no manual copy-pasting. Setting up the official server is covered in the GitHub MCP guide.

Databases (PostgreSQL, SQLite)

Probably the most underrated use case. Connect a server to a database, and you can ask questions about your data in plain language instead of writing SQL by hand. The agent figures out the schema, builds the query, and returns a result. How to set this up is in the separate article on connecting a database through MCP. For a specific engine, there's a step-by-step guide for PostgreSQL.

The agent reads messages from chats and channels and can send them too. Useful for summaries ("pull together what was said in the work channel today") and for scheduled tasks. Step by step, in the article on connecting Telegram through MCP.

Notion and knowledge bases

Through the official Notion MCP, an agent searches a workspace, pulls together a report from several pages, and updates documents. Connection happens through OAuth, so there's no password or API token to copy by hand.

Browser

Playwright MCP is for when reading a page isn't enough: the agent needs to click buttons, fill in fields, and check the state of an interface. It's overkill for regular search, but exactly right for verifying a user flow.

Calendar and email

Access to a schedule and inbox: find tomorrow's meetings, pull up a specific email, draft a reply. Paired with schedules, this turns into a background assistant that gathers what came in overnight by morning.

◆

Anything already built into Doka doesn't need its own MCP server: files, web search, page reading, and the terminal work out of the box. MCP is specifically for external services — things an agent doesn't have by default.

How to add a server in Doka

The mechanics are the same for any MCP server:

Open the right panel → the "MCP Servers" tab → "+ Add server."
Enter a name, the launch command, and its arguments (one per line) — usually taken from that specific server's README.
If the server needs a token or access key, add it under environment variables as KEY=VALUE.
Flip the toggle and save. The agent sees the new tools on the next task.

Every MCP server in Doka runs locally, as a process on your machine — data moves directly between the agent and the service, with no cloud in between.

Where to start

Don't connect everything at once. Pick one server for a real task — a database or Telegram, say — and see how the agent works with it. Once you've got the hang of one, the rest follow the same pattern.

The same list is useful in other hosts too. How servers connect in LM Studio, and why some of them overload a local model, is covered in the article on MCP in LM Studio. There's also a breakdown of experimental support for the protocol right inside the engine: MCP in llama.cpp.

A separate case is when the MCP server lives not in a standalone program but inside the app itself — that's how it's done in Unreal Engine 5.8, where the editor exposes its own operations to the agent directly.

If you haven't installed Doka yet — download it for free, MCP is already inside.

Useful MCP Servers — What to Connect to an AI Agent