Wiki · Start here

MCP tools

Ready-made interfaces the agent plugs into — suddenly it works with your systems, with no code written.

✅ Verified 15. 10. 2026Source: official documentation →📋 Copy for your AI
🔌
Many services already speak a language agents understand — your agent just connects. It is called MCP, and it is how you give the agent abilities it does not have on its own: access to your files, to a GitHub project, to your automation tool or to a browser. No code is written for it.

What it is, in plain words

On its own the agent works with a terminal, files and the web. Making it handle everything else would mean a custom solution per service — slow and fragile. MCP is a shared agreement that many services already offer. The agent connects to that ready-made interface and can immediately do what it offers.

What it gives youExample
The agent can see into your systemsWalks through project files, reads documentation, looks into a database
It drives tools you already useCreates a card in your automation tool, comments on a task, uploads a file
It can look at the web like a personOpens a page in a browser and verifies it really looks right
It reaches errors and documentationReads reported errors from your monitoring, looks up current library docs

Two kinds of connections

KindHow it worksTypically for
LocalA program runs on your own server; the agent talks to it inside the machineAccess to folders, a database, a local tool
RemoteThe agent connects over the internet to a service; some ask you to sign inGitHub, project management, error tracking

How to add a connection

  1. From the catalog — the easy route. Selected connections have a catalog and install with one command:
    $ hermes mcp catalog $ hermes mcp install
    For some connections you pick which tools the agent should see.
  2. By adding an address or command. When a connection is not in the catalog:
    $ hermes mcp add
  3. By editing the settings. The manual route, for when you know what you are doing — in ~/.hermes/config.yaml:
    mcp_servers: filesystem: command: "npx" args: ["-y", "@modelcontextprotocol/server-filesystem", "/home/user/projects"]

Connections that require a sign-in use hermes mcp login — you log in through the browser, just like any other account. Keys are not written into the settings; a connection reads them from the environment.

Verification: hermes mcp test tells you whether the connection came up and which tools it brought. Changes take effect after the agent restarts.

Safety: a connection is third-party code

⚠️
A connection gives the agent new abilities — and it can do essentially anything it is capable of. Only use sources you trust (the official catalog, verified vendors), and for connections that offer it, limit which tools the agent can see. A connection from an unknown source is like running a stranger's program with access to your server.
SafeguardWhat it does
The official catalogCatalog connections have been reviewed, so it is the safe place to start
Tool selectionYou can tick only what you actually want — the agent never even sees the rest
Keys kept out of the fileSecrets come from the environment, not from configuration, so they are not copied into backups or git

When it does not work

SymptomCause and what to do
The agent does not see the new toolsThe change applies after the agent restarts. Then verify with hermes mcp test
The connection reports a sign-in errorRun hermes mcp login and sign in through the browser; note that a sign-in does expire
A particular tool is missingIt was left out of the tool selection — hermes mcp configure adjusts it
The agent confuses the toolsThe connection brought too many tools. Narrow the selection to the ones you use
I do not know what is connectedAsk in the conversation; in the settings it is the mcp_servers section

Our tips

TipWhy
Start with the catalogNothing to invent, and you know the connection was reviewed
Add them one at a timeEnabling ten at once hides which one misbehaves and drowns the agent in tools
Narrow the tool selectionFewer tools means less confusion and less risk of the agent reaching for something you did not intend
Disconnect what you do not useEvery connection is more third-party code on your server

Where next

← Back to the wiki