Skip to content
DEVELOPERSThe essentials in 2 min

Connect an MCP client

One server address for the tools in your toolset.

The outcome

The client lists and calls the tools exposed by Vela.

✦Your AI agentMCP
Vela toolset
↗find_customerREAD
↗get_available_slotsREAD
↗read_order_statusREAD
Workflow illustration · example data

Your path, in three steps.

  1. 01

    Copy the MCP configuration

    The toolset page provides the server address and configuration to adapt to your client.

  2. 02

    Configure authentication

    Use the toolset token with Authorization: Bearer. This token differs from an individual connector key.

  3. 03

    Check exposed tools

    Check the tool list in the client and start with a read. Toolset restrictions still apply.

IN VELA · CREATE A TOOLSETEnlarge ↗
Create New MCP Toolset dialog: name, description and Read-Only Server option.
Read-only mode limits the tools available to your agent.Product screenshot · English interface.

Want to go deeper?

Open just the topic you need.

Overview

A Toolset is an MCP server over Streamable HTTP. If you are using Claude Desktop or Cursor, the AI assistants page is enough — this is for wiring it into something you wrote yourself.

✦Your AI agentMCP
Vela toolset
↗find_customerREAD
↗get_available_slotsREAD
↗read_order_statusREAD
One group, one revocable token: your assistant sees the tools, never your credentials.
Connecting
Client configuration
{
  "mcpServers": {
    "vela": {
      "url": "https://api.example.com/mcp/{toolsetId}",
      "headers": { "Authorization": "Bearer vela_ts_…" }
    }
  }
}

The two failures everyone hits

The header value must start with Bearer — without it no token is read and every call is refused with a 401. And the URL is the API host, not the dashboard host: the address in your browser bar answers 404 and explains nothing.

Methods
MethodBehaviour
initializeNegotiates the protocol version, advertises the tools capability.
tools/listThe enabled, live connectors of this Toolset. Input schemas come from each connector's own contract.
tools/callRuns one. A connector that refused returns a tool result with isError: true rather than a JSON-RPC error — the model is what has to react to a wrong date.
pingAnswers {}.
notifications/*202, empty body.
Annotations
AnnotationReading connectorActing connector
readOnlyHinttruefalse
destructiveHintfalsetrue
idempotentHintfalsefalse — a booking placed twice is two bookings
openWorldHinttruetrue

These are hints your client may ignore. The enforceable control is the Toolset's read-only switch, which hides acting connectors from tools/list and refuses them in tools/call — the second matters because an agent may be working from a list it fetched earlier.

What a call actually does

It runs the connector through the same path your own code would use, so it keeps the rate limit, the session handling, the execution history and the automatic repair. An agent call is an execution and is billed as one.

Tokens
  • The Toolset token is not a connector key. Revoking it breaks no direct integration.
  • It is stored hashed and can be rotated; the old one stops working immediately.
  • A wrong token and an unknown Toolset return the same message, so the endpoint cannot be used to discover which Toolsets exist.
UP NEXTGive an AI agent tools

Group the right actions into an MCP toolset.

Your situation doesn’t match the guide?

Talk to the team ↗