Connect Claude Desktop
Claude Desktop launches MCP servers as local processes, so it connects through
the first-party stdio bridge @evinor/mcp. The
bridge forwards every call to the hosted endpoint with your key. It runs with
npx, so Node.js must be installed.
The fastest way: Developers → Agent setup
In the Evinor web app, open Developers → Agent setup
(/agent-setup), pick your agent, and create the key. Evinor creates a dedicated API key that expires after 90 days and
builds the setup text below with that key already filled in. Copy it and paste
it into your agent, or follow the steps yourself.
By default the key gets every scope your account is allowed to grant. Open Adjust access before creating it to narrow that down. The scopes are fixed once the key exists.
The raw key is shown only while that page stays open. If you leave and come back, you create a new key. Each key uses one of the ten active-key slots on your account until you revoke it, even after it expires.
Set it up yourself
-
Open
claude_desktop_config.json(Claude Desktop → Settings → Developer → Edit Config) and merge thisevinorentry intomcpServers, with your ownevnr_live_…key:{
"mcpServers": {
"evinor": {
"command": "npx",
"args": ["-y", "@evinor/mcp"],
"env": {
"EVINOR_API_KEY": "evnr_live_…"
}
}
}
} -
Restart Claude Desktop, then check that the Evinor tools are available.
The bridge reads the key only from EVINOR_API_KEY; there is no command-line
flag for it. See Local (stdio) clients for the
other variables it reads.
Keep the key safe
- The key now sits in your agent's config and conversation history. Treat that file like a password: keep it out of version control and shared folders.
- Revoke the key under Developers → API keys when you stop using the agent, or as soon as you think it leaked. An expired key is refused but still holds a slot until you revoke it.
- Use a separate key per agent or machine, so you can revoke one without breaking the others.
Before you let the agent create sensors
A key with SENSORS_WRITE lets the agent create sensors. A sensor sends every
matching event to the webhook URL it was created with, and the agent chooses
that URL. Text inside search results comes from news coverage, so it could try
to steer the agent towards a URL you do not control. Check the destination URL
whenever your agent asks to create a sensor, or leave SENSORS_WRITE off the
key. See The sensor tools.
What to do next
Once the tools are listed, ask the agent what you want: search recent events,
follow a topic as a beat, or set up a sensor. The server tells the agent how
sensors, beats and credits work when it connects. Every search_events call
is charged; see Billing.
For the full tool list, scopes and errors, see MCP server.