With MCP, AI agents such as Claude or ChatGPT can use your team's FlowData services directly, for example to search for an address or check an email address. The server address is:
https://mcp.flowconnect.one/flowdata
Enter it exactly like this, with no slash at the end. FlowData has its own address and is not part of the FlowConnect bundle on mcp.flowconnect.one. To connect, you need access to FlowData in your team.
Connect by signing in (claude.ai, ChatGPT)
You connect claude.ai and ChatGPT without a key:
- Add a custom connector in claude.ai or ChatGPT.
- Enter the address above.
- Sign in with your account, choose your team and confirm the connection.
The connection then appears in FlowConnect under Developers › MCP, marked Connected by sign-in, and you can disconnect it there.
Connect with a key (Claude Code, Cursor and others)
Clients that accept a key use an MCP key. FlowData keys starting with fd_ do not work here.
- In FlowConnect, open Developers › MCP and create a new key.
- Give it a name and tick FlowData under Servers.
- Copy the key right away. It is only shown once.
- Enter the address and the key in your client:
{
"mcpServers": {
"flowdata": {
"type": "http",
"url": "https://mcp.flowconnect.one/flowdata",
"headers": { "Authorization": "Bearer mcp_YOUR_KEY" }
}
}
}
More on managing keys, such as revoking them, is in the FlowConnect article on API and MCP keys.
What the agent can do
| Area | Tool | Task |
|---|---|---|
| IP Geolocation | flowdata_geolocation |
Location data for an IP address or domain |
| Client Info | flowdata_client_info |
Browser, device and languages of the calling client |
| Maps | flowdata_maps_search |
Search for addresses and places |
| Maps | flowdata_maps_geocode |
Turn an address into coordinates or the other way round |
| Maps | flowdata_maps_directions |
Calculate a route between two points |
| Email Verifier | flowdata_email_verification |
Check an email address |
| YouTube | flowdata_youtube_stats |
Fetch channel statistics |
| Media | flowdata_media_create_job |
Create a media job |
| Media | flowdata_media_job_status |
Read the status and download links of a job |
With Client Info, the calling client is the MCP client, not your end customer.
In media processing, the agent can only create jobs and read their status. The source must be a publicly reachable https URL, and direct uploads and your own storage targets are not possible. Every output is stored by FlowData and deleted after seven days. Finished outputs carry a temporary download link.
Costs and responsibility
Every call is a normal, billed request on your team's FlowData account. Each paid tool's description gives the agent the list price, and your actual prices are under Billing. As with the API, errors, a Maps search with no results and the email result unknown are not charged, see How are FlowData prices calculated?.
You are responsible for everything your agent does in FlowData. The main points are:
- Set a spend limit so that an agent does not spend more than you want.
- Triggering
flowdata_media_create_jobagain creates a second job. Only start media jobs on explicit request. - If credit or the limit does not cover a call, FlowData rejects it, with HTTP 402 in the data areas and HTTP 429 for media.
The terms of the respective provider also apply when you connect AI agents.