Mida MCP Server is an official Model Context Protocol (MCP) integration that lets AI assistants like Claude, ChatGPT, Cursor, Copilot, and Gemini create, edit, preview, and analyze Mida A/B tests on your behalf — using a natural-language chat instead of the Mida dashboard. Once connected at https://mcp.mida.so/mcp, you can ask your AI assistant to spin up an experiment, set targeting rules, define goals, or summarize results, and Mida MCP executes those actions inside your Mida project automatically by calling the Mida REST API.
Prefer to watch? Watch the full video tutorial on YouTube — or jump straight to the section you need:
This guide covers the same workflow in writing.
What is the Mida MCP Server?
The Mida MCP Server is a hosted endpoint — https://mcp.mida.so/mcp — that exposes Mida's A/B testing platform to any MCP-compatible AI assistant. MCP — short for Model Context Protocol — is an open standard for connecting AI assistants to the tools and services you already use.
Mida itself is an A/B testing platform that runs on a lightweight JavaScript snippet installed on your site. You define experiments, variants, goals (clicks, form submissions, page views), and targeting rules, and Mida splits traffic and measures conversion lift. Everything you can do in Mida is also exposed via the Mida REST API — and the Mida MCP Server wraps that API so an AI assistant can drive it for you.
In practical terms, the Mida MCP Server allows you to:
Talk to your AI assistant (Claude, ChatGPT, or any MCP-compatible client) instead of clicking through the Mida UI.
Get experiments done automatically — the AI calls Mida's REST API in the background to create or modify tests.
Stay inside your AI workflow — no tab-switching, no context loss, no manual data entry.
If you can describe an A/B test in a sentence, the Mida MCP Server can build it.
Why use Mida MCP instead of the Mida dashboard?
Task | Without MCP | With Mida MCP |
Create a headline A/B test | Log in → pick page → write variant → set goal → configure targeting → save | One chat message describing the test |
Target Google Ads visitors only | Manually configure URL parameter rule | AI infers |
Summarize results across 10 tests | Open each test report individually | One prompt: "Summarize all live experiments" |
Find insights in a test report | Read charts manually | Ask AI to surface statistically significant findings |
The MCP layer doesn't replace the Mida dashboard — it gives you a second, conversational way to use it.
What can Mida MCP do?
Mida MCP exposes 20 tools to your AI assistant. You can configure each one to either run automatically or require your approval before execution.
Experiments
create_experiment— build a new A/B test (page, variants, targeting, goal)get_experiment— fetch a single experiment's full configurationlist_experiments— list all experiments in the current projectupdate_experiment_status— move an experiment between draft, live, paused, or archived
Results & analysis
get_experiment_result— pull the final test reportget_experiment_metrics— pull raw conversion metricsget_experiment_timeseries— pull conversion data over timecompute_experiment_statistics— calculate statistical significance and lift
Goals
create_goal— define a conversion goal (page view, click, custom event)list_goals— list all goals in the projectdelete_goal— remove a goal
Events
create_event— register a custom conversion eventlist_events— list all custom events in the project
Organizations, projects & context
list_organizations— list Mida organizations the connected account belongs toselect_organization— switch the MCP's active organizationlist_projects— list Mida projects in the active organizationselect_project— switch the MCP's active projectget_project_configuration— read project-level settings (integrations, defaults)get_mida_api_context— surface Mida API context to the AI assistant
Previews
generate_preview_url— produce a shareable preview link for a variant, with optional URL parameters
Some actions (like update_experiment_status or delete_goal) are good candidates to require approval; read-only tools like list_experiments or get_experiment_result can safely be set to Always allow for a hands-off workflow.
Prerequisites
Before you start, make sure you have:
A Mida account with at least one project, and the Mida JS snippet installed on the site you want to test. (Sign up at mida.so)
An MCP-compatible AI assistant. Mida officially supports ChatGPT, Claude Chat / Claude Desktop, Claude Code, Cursor, VS Code / Copilot, Gemini CLI, and Windsurf — plus any other compatible remote MCP client.
The Mida MCP Server URL:
https://mcp.mida.so/mcp— the same URL for every AI assistant.
How do I set up Mida MCP with Claude? (Step-by-step)
Step 1 — Get the Mida MCP Server URL
The Mida MCP Server URL is the same for every AI assistant:
https://mcp.mida.so/mcp
You can also find it in your Mida dashboard → Integrations → MCP Connections, where you'll see a copy button next to the URL alongside setup instructions for each supported assistant.
Step 2 — Add Mida as a custom connector in Claude
Open Claude (claude.ai or the desktop app).
Go to Customize → Connectors → Add custom connector.
Paste
https://mcp.mida.so/mcpas the MCP server URL.Give the connector a name — e.g.
Mida A/B Test.Save.
Step 3 — Authenticate and pick your project
Click Connect on the Mida connector inside Claude.
Claude opens a Mida login window — sign in with the Mida account you want the AI to control.
Select the specific Mida project you want the MCP to operate on.
Approve the connection.
Step 4 — Configure permissions
After connecting, Claude shows you all 20 Mida MCP tools under Tool permissions ("Choose when Claude is allowed to use these tools"). For each one, choose:
Always allow — AI executes without asking. Best for read-only tools (
list_experiments,get_experiment_result,get_experiment_metrics).Ask each time — AI requests your approval before executing. Best for destructive or live-impact actions (
update_experiment_status,delete_goal).
For a fully autonomous workflow, set all 20 actions to Always allow.
You're done. Mida MCP is now connected to Claude.
Setting up with a different assistant? The process is the same for ChatGPT, Cursor, VS Code / Copilot, Gemini CLI, Windsurf, and any other MCP-compatible client — same URL (https://mcp.mida.so/mcp), same auth flow, same 20 tools. Just add it as a custom connector inside that assistant. The Mida dashboard's Integrations → MCP Connections page has specific setup instructions for each assistant.
How do I create an A/B test with Mida MCP and Claude?
Here's the exact workflow shown in the video tutorial — a headline test on a landing page targeted at Google Ads visitors.
Step 1 — Describe the test in plain English
Open Claude (with the Mida connector enabled) and write a prompt that covers four things:
Which page you want to test
What change you want to test
What goal measures success
Who to target
Example prompt:
Create a Mida A/B test on www.mida.so.
Test changing the headline to "Modern A/B testing for marketers and CRO teams".
The goal is visitors landing on our signup page (page-view goal).
Target only visitors coming from Google Ads.
Step 2 — Let Mida MCP build the experiment
Claude calls Mida MCP, which creates the experiment as a draft. You'll get back:
The experiment name
The test ID
Control and Variant details
Targeting rules
Goal configuration
Step 3 — Preview the variant
The test is in Draft — preview it before going live.
If your targeting uses a URL parameter (like GCLID), the default preview URL won't show the variant because it doesn't carry that parameter. Just ask Claude:
Add the Google Ads GCLID parameter to the preview URL so I can see the variant.
Claude returns a preview link with the parameter appended. Open it, and the new headline appears.
Step 4 — Set the experiment live
Once the preview looks right, tell Claude:
Looks great — set the experiment live.
Mida MCP updates the experiment status to Live. Traffic starts splitting (50/50 by default, 95% confidence level, all project-level integrations auto-enabled).
Step 5 — Verify on the Mida dashboard (optional)
Log in to Mida → open the experiment. You should see:
The landing page URL you specified
The change as the variant
The targeting rule you specified
Your experiment goal
50/50 traffic split, 95% confidence
What other A/B testing workflows can I run through Mida MCP?
Anything you can do in the Mida dashboard, you can do by chatting with your AI assistant. Common workflows:
"Summarize all my live experiments and flag any that have reached significance."
"Look at the test report for [experiment ID] and tell me what's working."
"Create a button-color test on the pricing page with a click goal."
"Pause every experiment that's been running for more than 30 days without a winner."
"Build the same headline test, but target mobile visitors from organic search."
"What's my best-performing experiment this month?"
FAQ
What is the Mida MCP Server?
The Mida MCP Server is Mida's official Model Context Protocol integration. It lets AI assistants like Claude or ChatGPT create, edit, preview, and analyze Mida A/B tests through natural-language chat instead of the Mida dashboard.
What is MCP?
MCP (Model Context Protocol) is an open standard for connecting AI assistants to external tools and services. It lets an AI assistant call real actions in apps you already use — like creating an A/B test in Mida — without you logging into those apps manually.
Which AI assistants work with Mida MCP?
Mida MCP officially supports ChatGPT, Claude Chat / Claude Desktop, Claude Code, Cursor, VS Code / Copilot, Gemini CLI, and Windsurf — plus any other MCP-compatible client that supports remote MCP servers. You'll find copy-ready setup links for each one on the Integrations → MCP Connections page in your Mida dashboard.
Do I need to be a developer to use Mida MCP?
No. Setup is point-and-click: copy the MCP URL from Mida, paste it into Claude or ChatGPT as a custom connector, log in, and pick a project. No code, no API keys to manage manually.
Can Mida MCP create A/B tests automatically?
Yes. Describe the test in a sentence — what page, what change, what goal, who to target — and Mida MCP creates the experiment as a draft. You preview it, approve it, and set it live, all from the same chat.
Does Mida MCP support Google Ads targeting?
Yes. If you tell the AI to "target Google Ads visitors," Mida MCP automatically configures targeting on the GCLID URL parameter, which Google Ads appends to every ad click.
Can I preview an A/B test variant before going live?
Yes. After Mida MCP creates the experiment, ask your AI assistant for the variant preview URL. If your targeting uses URL parameters (like GCLID), tell the AI to include those parameters in the preview link.
Can the AI set my experiments live without asking?
Optionally — yes. When you connect Mida MCP, you choose which actions require your approval and which run automatically. Set "update experiment status" to Always allow if you want fully hands-off operation.
What actions does Mida MCP support?
Mida MCP currently exposes 20 tools: create_experiment, get_experiment, list_experiments, update_experiment_status, get_experiment_result, get_experiment_metrics, get_experiment_timeseries, compute_experiment_statistics, create_goal, list_goals, delete_goal, create_event, list_events, list_organizations, select_organization, list_projects, select_project, get_project_configuration, get_mida_api_context, and generate_preview_url.
How is Mida MCP different from the Mida REST API?
The Mida REST API is a developer interface that requires you to write code, manage a Project Key and API Key, and pick a US or EU regional endpoint. Mida MCP wraps that API so an AI assistant can call it on your behalf — you just chat. Both interfaces hit the same underlying platform, so anything created via MCP appears in the dashboard and the API immediately, and vice versa.
How much does Mida MCP cost?
Mida MCP is included with your existing Mida plan at no extra cost. You pay your normal AI assistant subscription (Claude or ChatGPT) separately.
Is Mida MCP secure?
Yes. Mida MCP authenticates through your Mida login (OAuth), is scoped to the specific project you select during setup, and lets you require explicit approval for sensitive actions. No long-lived API keys are stored in the AI assistant.
What if I want to switch between Mida organizations or projects?
Just ask your AI assistant to "switch to organization Y" or "switch to project X" — it will call list_organizations / select_organization or list_projects / select_project for you. No need to reconnect or re-authenticate.
Can Mida MCP analyze experiment results?
Yes. Ask your AI assistant to summarize all live experiments, look at a specific test report, or surface statistically significant findings. Mida MCP pulls the metrics and the AI handles the narrative summary.
Troubleshooting
The connector says "already connected" when I try to add it.
You've already added Mida MCP to this AI assistant. Disconnect the existing connector first, then re-authenticate with your Mida account.
The variant preview doesn't show my change.
Your experiment targeting probably uses a URL parameter (e.g. GCLID for Google Ads). Ask the AI to append the parameter to the preview URL, then reload.
The AI says it doesn't have permission to set the experiment live.
Open your AI assistant's connector settings → Mida → permissions, and change "Update experiment status" from Ask each time to Always allow (or approve the action when prompted).
Claude can see the connector but can't reach Mida.
Re-check the MCP server URL — it should be exactly https://mcp.mida.so/mcp. If it still fails, disconnect, re-add the connector, and re-authenticate with your Mida account.
Summary
Mida MCP turns A/B testing into a conversation. Connect once, then ask Claude, ChatGPT, or any MCP-compatible AI assistant to build experiments, manage targeting, preview variants, and summarize results — without ever opening the Mida dashboard. Setup takes under five minutes, and your first AI-built experiment can be live in another five.
If you have questions, reach out to the Mida team — we're happy to help.