Skip to main content
MCP (Model Context Protocol) lets you use Markifact’s operations directly inside your favorite AI client — Claude, ChatGPT, Gemini, Manus, Cursor, and any tool that supports MCP. Instead of building workflows or chatting with the Markifact agent, you stay in the AI tool you already use and ask it to pull reports, create campaigns, send emails, and more — all powered by Markifact behind the scenes.

​
What is MCP?

MCP is an open standard that lets AI clients connect to external tools and services. When you create an MCP server in Markifact, you’re giving your AI client access to Markifact’s 300+ marketing operations. Your AI client can then:
  • Search for available operations
  • Pull reports from Google Ads, Meta Ads, GA4, and more
  • Create and manage campaigns
  • Send data to Sheets, Slides, Slack, and other destinations
  • Work with files and data across platforms
All without leaving the AI conversation.

​
MCP vs Workflows vs Agents

Markifact gives you three ways to get things done. Each is suited to different scenarios:
WorkflowsAgentsMCP
What it isAutomated sequences of steps that run on a schedule or triggerA conversational AI inside Markifact that executes operations for youA bridge that lets external AI clients use Markifact operations
Best forRecurring, repeatable tasks (e.g. weekly reports, daily syncs)One-off tasks and exploration inside MarkifactUsing Markifact from your preferred AI tool (Claude, ChatGPT, Cursor, etc.)
How it runsAutomatically on a schedule, trigger, or manuallyYou chat and the agent picks the right operationsYour AI client calls Markifact operations during a conversation
SetupBuild once on the canvas, runs foreverJust start a conversationCreate a server, add a token, paste the config into your AI client

​
When to use what

  • Use Workflows when you have a repeatable task that should run automatically — like sending a weekly Google Ads report to Slides every Monday.
  • Use Agents when you want to do something ad-hoc inside Markifact — like “pull last week’s Meta Ads performance and summarize it.”
  • Use MCP when you want to stay in your AI client and use Markifact’s capabilities from there — like asking Claude to pull your GA4 data and compare it with last month.

​
Getting Started

​
1. Create an MCP Server

  1. Go to the MCP page from the sidebar.
  2. Click New MCP Server.
  3. Give it a name (e.g. “Claude”, “Manus”).

​
2. Connect Your AI Client

There are two ways to authenticate your AI client with Markifact: OAuth and Manual Token. The method you use depends on what your AI client supports. OAuth is the most secure and convenient way to connect. If your AI client supports OAuth — like Claude — you only need to paste the MCP server URL: 
https://api.markifact.com/mcp
When you connect, your AI client will redirect you to Markifact where you authorize access. No tokens to copy or manage — everything is handled automatically. This is the recommended method whenever your AI client supports it.

​
Manual Token

For AI clients that don’t support OAuth (like Cursor or other developer tools), you can use a manual token instead:
  1. On your MCP server settings page, a token is automatically generated for you.
  2. Copy the token and use it in your AI client’s MCP configuration:
{
  "mcpServers": {
    "markifact": {
      "url": "https://api.markifact.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN_HERE"
      }
    }
  }
}
Paste this into your AI client’s MCP settings (the exact location depends on the client — check their docs for where to add MCP servers).

​
Permissions & Security

Each MCP server can be configured with fine-grained permissions to control exactly what your AI client can access.

​
Operations

By default, a server has access to all operations. You can restrict it to specific operations only — for example, allow only Google Ads reporting, or only Sheets operations.

​
Connections

By default, a server can use all connected accounts. You can restrict it to specific connections — useful when you want a server that only accesses one client’s Google Ads account, for example. This makes it safe to create dedicated servers for different purposes or share access with team members without exposing everything.

​
Tool Permissions in Your AI Client

Most AI clients let you control when each tool is allowed to run. We recommend the following setup:
  • Always allow all read-only tools — find_operations, get_operation_inputs, list_connections, get_file_url, read_file, and run_operation. These only read data and are safe to auto-approve.
  • Require approval for run_write_operation. This tool creates, updates, or deletes data on external platforms. Keeping it on “Needs approval” lets you review exactly what the AI is about to do before it executes.
Do not set run_write_operation to “Always allow.” Write operations can modify live campaigns, send emails, update budgets, and make other irreversible changes. Always review what the AI is requesting before approving a write action.
In Claude, for example, this looks like setting Read-only tools to “Always allow” and Write/delete tools to “Needs approval” — which is the default.

​
Available Tools

When your AI client connects to Markifact via MCP, it gets access to the following tools:

​
find_operations

Search for available operations across all platforms. This is the starting point — your AI client uses this to discover what’s available. For example, searching “google ads report” returns matching operations with their IDs and descriptions.

​
get_operation_inputs

Get the input schema for a specific operation. This tells the AI client what fields are required, what the defaults are, and what values are accepted — so it can fill in the right parameters before running an operation.

​
run_operation

Execute a read-only operation. This covers things like pulling reports, listing accounts, searching for metrics, reading data from Sheets, and similar non-destructive actions.

​
run_write_operation

Execute an operation that creates, updates, or deletes data. This includes actions like creating campaigns, updating ad budgets, sending emails, or modifying Sheets data. Write operations are separated from read operations as a safety measure. Before running a write operation, the AI client is instructed to describe exactly what will change and ask for your confirmation first. This ensures you always stay in control of destructive actions.

​
list_connections

View which platform accounts are connected (e.g. Google Ads, Meta Ads, Slack). Useful when you have multiple accounts for the same platform and need to specify which one to use. Most of the time connections are resolved automatically — this tool is only needed when disambiguation is required.

​
get_file_url

Get a shareable link for a file produced by an operation. Some operations return files (like exported CSVs or generated charts) — this tool creates a temporary signed URL you can open or share.

​
read_file

Inspect the contents of a file returned by a previous operation. Useful for reviewing data before passing it to another operation — like checking the structure of a CSV export before sending it to Sheets.

​
How Credits Work

MCP credit usage is straightforward:
  • Discovery and utility tools are free. find_operations, get_operation_inputs, list_connections, get_file_url, and read_file cost 0 credits.
  • Running operations costs credits, and the cost depends on the specific operation — the same way workflow nodes are priced.

​
Examples

ActionCredits
Search for operations (find_operations)0
Get operation schema (get_operation_inputs)0
List connections (list_connections)0
Pull a Google Ads report (gads_get_report)1
Pull a GA4 report (ga4_get_report)1
Select accounts (gads_select_accounts)0
List report fields (ga4_list_report_fields)0
Read data from Sheets (sheets_read_data)1
Create a Google Ads campaign (gads_create_campaign)1
Send an email (email_send)1
The rule of thumb: utility and discovery operations (account selection, listing fields, searching) are free. Data operations (pulling reports, reading/writing to platforms) cost credits, typically 1 credit per operation, same as the equivalent workflow node. You can check exact credit costs for any operation by looking at the corresponding node documentation, or by checking the credit estimate shown in the workflow editor.

​
Tracking MCP Usage

MCP credit consumption appears in the Usage Dashboard under the MCP service type. Each operation execution is logged with the server name and operation ID so you can track exactly where credits are going.

​
Token Management

For manual token authentication, each MCP server can have one active token at a time. You can:
  • Rotate a token — generates a new token and revokes the old one. Use this if a token is compromised or you want to cycle credentials.
  • Copy the token — to paste into a new AI client configuration.
Tokens are shown once when created. If you lose a token, simply rotate it to generate a new one. If you’re using OAuth, token management is handled automatically — there’s nothing to copy or rotate.

​
Frequently Asked Questions

Yes. Create as many as you need — one per AI client, one per team member, or one per use case. Each server has its own token and permissions.
No. MCP servers use the same connections you’ve already set up in Markifact. If you have Google Ads connected in your workspace, your MCP server can use it automatically.
Use OAuth whenever your AI client supports it — it’s more secure and easier to manage. Use a manual token only when your AI client doesn’t support OAuth (like Cursor or other developer tools).
All tokens for that server are immediately revoked. Any AI client using those tokens will lose access. OAuth sessions are also invalidated.
There’s no rate limit specific to MCP. Operations consume credits the same way as workflows, and your usage is limited by your plan’s credit balance.
Check your server permissions:
  • If you restricted the server to specific operations, make sure the ones you need are included.
  • Go to your MCP server settings and check the Operations section.
Check your connection:
  • Make sure the MCP server URL and token are correct in your AI client.
  • Try regenerating the token if using manual authentication.
This usually means you haven’t connected the required platform account in Markifact yet.How to fix:
  1. Go to the Connections page in Markifact.
  2. Connect the platform account your operation needs (e.g. Google Ads, Meta Ads).
  3. Try the operation again — connections are resolved automatically.
If you have multiple accounts for the same platform, you may need to specify which one to use via the list_connections tool.