Quickstart - Everflow API

EU Cluster Support: If your Everflow account is hosted in the EU, simply replace https://mcp.eflow.team with https://mcp-eu.eflow.team in any of the configuration examples below.

Prerequisites

A Network API key with access to your network. See Authentication.

Connect your client

Each setup takes under 5 minutes. Depending on where your account is hosted, choose the appropriate server URL:

US-Hosted Accounts: https://mcp.eflow.team
EU-Hosted Accounts: https://mcp-eu.eflow.team

All clients authenticate with an API-key header — only the config format varies.

Two header names work, everywhere. The MCP server accepts X-Eflow-API-Key and X-Api-Key interchangeably, on every endpoint and in every client — this is not a per-client or Slack-only alias. The examples below use X-Eflow-API-Key; substitute X-Api-Key freely, and use it wherever a client only accepts standard header names. (This alias is MCP-only — the REST API at api.eflow.team still requires X-Eflow-API-Key.)

If you’re unsure where to start, use Claude.ai — the connector UI needs no config file, and it’s the only path that also gives you Slack access and scheduled tasks.

Open your Claude Desktop config file

macOS

open ~/Library/Application\ Support/Claude/claude_desktop_config.json

Windows

notepad %APPDATA%\Claude\claude_desktop_config.json

Add the Everflow server to the mcpServers block

{
  "mcpServers": {
    "everflow": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.eflow.team",
        "--header",
        "X-Eflow-API-Key:your-network-api-key"
      ]
    }
  }
}

Claude Desktop currently requires the mcp-remote stdio bridge to connect to remote MCP servers. The npx -y prefix auto-installs the package on first launch — no separate install step is needed. For the latest on native remote MCP support in Claude Desktop, see Anthropic’s changelog.

The --header value uses Name:Value with no space after the colon. X-Eflow-API-Key: your-network-api-key (with a space) will fail.

Save the file and restart Claude Desktop

Quit and reopen the app so it picks up the new MCP server.

Test the connection

Open a new conversation. You should see an Everflow tools indicator in the input area. Ask:

“What is my network’s name and default currency?”

Claude Code is Anthropic’s CLI agent, with native support for remote MCP servers over HTTP.

Add the Everflow server

claude mcp add --transport http everflow https://mcp.eflow.team \
  --header "X-Eflow-API-Key: your-network-api-key"

Verify the connection

Run /mcp inside Claude Code to confirm the everflow server is connected, then ask:

“Call get_account_info and tell me my network name and default currency.”

Claude.ai now supports custom connectors with request headers, so you can connect Everflow directly — no mcp-remote bridge and no config file. This is also what powers Everflow answers in a Slack DM and scheduled reports.

Open the connector dialog

Go to Settings → Connectors → Add custom connector.

Name the connector and paste the server URL

Name it Everflow MCP, and set the URL to https://mcp.eflow.team (or https://mcp-eu.eflow.team for EU-hosted accounts).

Set Authentication to None

The dialog may pre-select Always required with a Detected badge. Override it and choose None — Everflow MCP authenticates with an API key, not OAuth. Leaving OAuth selected sends Claude into a sign-in flow that will fail.

Add the API key as a request header

Under Request headers, add a header named x-api-key, paste your Network API key as the value, and tick Required.

Use x-api-key here, not X-Eflow-API-Key. The constraint is Claude’s, not Everflow’s: Claude accepts the standard header names (authorization, x-api-key) straight away, while any other name needs Anthropic’s review before the connector can be saved. X-Api-Key works against the Everflow MCP server everywhere, not just here, so nothing about this setup is special-cased.

Add, then connect

Click Add, then Connect on the connector’s page. Everflow’s read tools appear under Tool permissions, plus the write tools if the key’s MCP access is Full access.

Choose how much approval you want

Each tool can be set to allow automatically, ask for approval, or block. Needs approval is the default and is a sensible starting point; switch the read tools you use constantly — run_performance_report, list_offers, list_affiliates — to allow once you trust the setup.

Don’t paste an admin or all-modules key here. The key is stored by Anthropic and never shown again, and its permissions become the connector’s entire capability surface. Create a dedicated Read Only key for this connector, granted only the modules it needs, with MCP access left at Read Only. A Reporting key alone can return consumer emails, IPs, and mobile ad IDs. See Choosing a key for a connector.

Open Cursor's MCP settings

Go to Cursor Settings → MCP, or press Cmd+Shift+P → MCP: Edit Configuration.

Add the Everflow server

{
  "mcpServers": {
    "everflow": {
      "url": "https://mcp.eflow.team",
      "headers": {
        "X-Eflow-API-Key": "your-network-api-key"
      }
    }
  }
}

This config holds your network API key in plaintext. If it lives in a workspace .cursor/mcp.json, add it to .gitignore so the key never lands in a committed repo.

Save and reload the window

The Everflow tools will appear in the Cursor Agent chat.

Gemini CLI is Google’s open-source terminal agent with native MCP support.

Install Gemini CLI

npm install -g @google/gemini-cli

Open the Gemini CLI config file

open ~/.gemini/settings.json

Create the file if it doesn’t exist yet.

Add the Everflow server

{
  "mcpServers": {
    "everflow": {
      "httpUrl": "https://mcp.eflow.team",
      "headers": {
        "X-Eflow-API-Key": "your-network-api-key"
      }
    }
  }
}

Start a session and verify the connection

gemini

“Call get_account_info and tell me my network name and default currency.”

VS Code requires an active GitHub Copilot Chat subscription and the Copilot Chat extension. The config format also differs from the other clients — note the servers key instead of mcpServers.

Install the GitHub Copilot Chat extension

Install it from the VS Code marketplace if you don’t have it already.

Open your MCP configuration

Open VS Code settings (Cmd+,) and search for MCP Servers, or edit .vscode/mcp.json directly in your workspace:

{
  "servers": {
    "everflow": {
      "url": "https://mcp.eflow.team",
      "headers": {
        "X-Eflow-API-Key": "your-network-api-key"
      }
    }
  }
}

.vscode/mcp.json holds your network API key in plaintext and often lives inside a repo. Add it to .gitignore so the key is never committed.

Open Copilot Chat in Agent mode

The Everflow tools will appear in the tool picker.

Verify the connection

Once connected in any client, run this prompt to confirm everything is working:

“Call get_account_info and tell me my network name, default timezone, and default currency.”

A successful response confirms your API key is valid and the MCP server is reachable. get_account_info needs the Control Center Accounts (Settings) permission, so it is hidden from a key without it, such as a Read Only Reporting key. On such a key, verify instead with run_network_summary for yesterday or list_offers with page_size=1, whichever module your key has.

First questions to try

Not sure where to start? Ask in plain language — a few examples:

“How did my top offers perform last week?”
“Why was transaction <id> rejected?”
“Which partners have the highest conversion rate but low volume?”
“Is offer <id> pacing toward its cap?”
“Show revenue by partner for this month.”

The server also advertises its capabilities to the assistant on connect, so “what can you help me with?” gives a grounded, Everflow-specific answer.

Debug with MCP Inspector

If a client isn’t connecting or tools aren’t appearing, the MCP Inspector lets you test the server directly — independent of any AI client config.

Run the inspector

No install required:

npx @modelcontextprotocol/inspector

Configure the connection in the inspector UI

Field	Value
Transport	Streamable HTTP
URL	`https://mcp.eflow.team`
Header name	`X-Eflow-API-Key`
Header value	`your-network-api-key`

Connect and call a test tool

Click Connect, then call get_account_info from the Tools tab. A valid response confirms the server is reachable and your key is accepted (on a key without the Accounts permission the tool is hidden, so call run_network_summary instead). If this works but your AI client still can’t connect, the issue is in your client config, not the server.

Troubleshooting

Symptom	Likely cause	Fix
Connection refused or timeout	Wrong server URL or a network/firewall issue	Confirm the URL (`https://mcp.eflow.team`, or `https://mcp-eu.eflow.team` for EU) and that outbound HTTPS is allowed
`401 Unauthorized`	Invalid or missing API key	Verify the key in Control Center → Security
`403 Forbidden`	Key is not a Network key	Affiliate and advertiser keys are not accepted here (partners use the Affiliate MCP)
`429 Too many requests`	Over 10 requests per second for the network	Retry after 1 second (see the `Retry-After` header); the limit is shared by every user and key on the network
Tools not appearing in client	Config file syntax error	Validate your JSON — trailing commas are invalid
`not valid MCP server configurations and were skipped` on Claude Desktop launch	Claude Desktop does not accept the `url` + `headers` (Streamable HTTP) format	Use the `mcp-remote` stdio config shown in the Claude Desktop tab above
Partial data returned	Limited affiliate scope on your key	Employees with scoped access only see their assigned affiliates
`npx` fails with “command not found” or hangs	Node.js not installed	Install Node.js from nodejs.org, then retry
A tool or action the docs describe is “unsupported”, or your client rejects an argument the docs list	Your client cached the tool list when it connected, and Everflow has released a change since	Reconnect: restart your MCP client (or its connection). See Staying current after a release
An argument is rejected before any Everflow error appears (e.g. a JSON-RPC `-32602`)	Same cause: the rejection is your client validating against its cached schema, not Everflow refusing the call	Reconnect, then retry
The write tools are missing, or a write call says the key’s MCP access is Read Only	The key’s MCP access is not Full access, or it lacks the tool’s permission	Change the key under Control Center → Security, then reconnect
Every request fails with “MCP access is turned off for this API key”	The key’s MCP access is Off	Set it to Read Only or Full access, or use another key

Staying current after a release

MCP clients fetch the tool list once, when they connect, and validate your arguments against that cached copy. Everflow’s tool set only changes when the server is redeployed, and a redeploy ends existing connections, so there is no in-session update: a long-lived client keeps using whatever contract it captured at connect time. In practice that means a client connected before a release will:

still advertise old actions and parameters, and describe removed ones as available;
reject new arguments locally, before the call reaches Everflow, usually as a JSON-RPC -32602 naming a parameter that the action you are calling does not even use.

Reconnect after an Everflow MCP release (restart the client, or its connection) and it will pick up the current tools. The same applies after you change a key’s permissions or its MCP access. If a documented capability appears to be missing, reconnect before investigating anything else.

What’s next

Prompt Library

Ready-to-use prompts for reporting, partner health checks, and traffic investigation.

Agentic Examples

End-to-end multi-step workflow traces showing exactly what the agent does.

​Prerequisites

​Connect your client

​Verify the connection

​First questions to try

​Debug with MCP Inspector

​Troubleshooting

​Staying current after a release

​What’s next

Prompt Library

Agentic Examples

Prerequisites

Connect your client

Verify the connection

First questions to try

Debug with MCP Inspector

Troubleshooting

Staying current after a release

What’s next