Connect an agent
How agents read Velur through our MCP server, what they can and cannot see, and how to connect yours.
What the MCP server is
Velur runs an MCP server at https://www.velur.io/mcp. MCP, the Model Context Protocol, is an open standard that lets an agent use the tools of another service with your permission. Claude, ChatGPT, Cursor, Codex, Gemini CLI and the other agents that speak it over HTTP with sign-in can read your Velur workspaces through it, and we do not build a separate integration for each one.
Server URL
https://www.velur.io/mcpHow an agent connects
You add the address to your agent once: as a connector in a chat app, as an MCP server in an editor, or with one command in your terminal. The first time the agent uses it, it opens our page in your browser. You sign in with your Velur account, pass two-step sign-in if your account has it, see which workspaces the agent will read, and press Allow. The agent then holds a token for your account, so there is no API key to copy, store or rotate.
What an agent can read
An agent reads through seven tools, and every one of them is read-only. Each tool runs the console's own code, so a figure an agent quotes is the figure on your screen and in your morning brief.
- Your workspaceslist_workspacesEvery store you can read, with its currency and timezone.
- One day's briefget_briefThe headline, the eleven figures and what was worth a look.
- Recent briefslist_briefsThe latest days, newest first.
- A metric, day by dayget_metric_historyOne figure across days, each with its trace id.
- A week, month or quarterget_period_reportA period against the one before, summed by our code.
- The trace behind a figuretrace_figureIts formula, its inputs and the sync it came from.
- Sources and freshnessget_sourcesWhen each source last synced, and the last complete day.
Every figure comes back finished, formatted in your store's currency, with its trace id. The agent is told to quote figures as given and to call a figure missing rather than work it out. We cannot stop another model from doing arithmetic, so check any figure it quotes against its trace id in the console.
What an agent can never do
It cannot change anything in Velur or in your store. It never sees a customer, an order, the people your brief goes to, the flags your team raises or a credential. The database refuses those tables to its token, whatever a tool asks for.
Connect yours
Every agent below has its own steps on velur.io/agents, with the commands and the config files to copy. Each one uses the same address and asks you to sign in and press Allow the first time.
- Antigravity App
- Antigravity CLI
- Antigravity IDE
- ChatGPT
- Claude Code CLI
- Claude Desktop and Web
- Codex app
- Codex CLI
- Cursor
- Devin
- Gemini CLI
- GitHub Copilot CLI
- Grok Bot
- Kimi Code CLI
- Manus
- VS Code
- Zed
- Other clients
Questions to start with
- Revenue was down yesterday. Was it fewer orders, smaller baskets or more refunds?
- Is our ad spend still paying back? Compare blended ROAS this week with last week.
- Where did our money go this month: discounts, refunds or ads? How does that compare with last month?
- Are we still winning new customers, or is the growth coming from people who already bought?
- Before I send these numbers to the team, is any source behind?
Who can see what
The agent reads the workspaces you can read, as you. Allowing it is recorded in the activity log of each of those workspaces, and the workspace owners get an email. To disconnect it, remove Velur from the agent's connectors or MCP settings.
