
Orphex
co.orphexPublished by orphex.coOfficial publisherv1.0.1Updated Nov 7, 2026
Analyze ad performance across platforms and execute changes after your approval.
Overview
Remote MCP server that lets an assistant analyze ad spend, ROAS and campaigns across ad platforms and apply changes only after your approval.
- What you can do
- Analyze performance across ad, analytics and commerce platforms: spend, ROAS, CPA, pacing and anomalies
- Build weekly or monthly performance reports and executive summaries from workspace templates
- Read live campaign status and budgets directly from a named platform
- Review change history, including who made each change
- Propose changes such as budgets, bids, campaign status, new campaigns or negative keywords, with a preview before applying
- When to use it
- Useful when you manage ad accounts in Orphex and want an assistant to explain why metrics moved, produce recurring performance reports, or make routine campaign changes with an explicit approval step.
- Requirements
- A remote Streamable HTTP endpoint; no local package. An Orphex account with at least one workspace, and platform connections set up in Orphex. Sign-in is OAuth in the browser, with no API key. Some clients need a free local port during sign-in (33418 for Claude Code, 7777 for Gemini CLI). Orphex only admits a known list of clients and does not accept dynamic client registration.
Installation
In SourceWeft
- Open Orphex in the dashboard and add it to a workspace.
- Connect the required credentials. They are stored privately and never shown on public pages.
- Enable the server for the chats that should use its tools.
Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.
Use with an agent
Paste this prompt into Claude Code, Codex, Cursor or another agent — or into SourceWeft chat. The agent reads this server's install guide, shows you its publisher, endpoint and sign-in, and adds it to your client once you agree. Keys never go into a command.
Read https://sourceweft.com/mcp/co.orphex%2Fmcp/install.md and connect the MCP server it describes to my client. Before changing any config, show me its publisher, endpoint and sign-in method and wait for my OK. Never put an API key in a command or in a file I would commit; ask me where it should come from.Other MCP clients
Add this to your client's mcpServers config. The client completes OAuth sign-in on first connect.
{
"mcpServers": {
"orphex": {
"type": "http",
"url": "https://api.orphex.co/api/interaction-gateway/v1/mcp"
}
}
}README
Orphex MCP
Ask your AI assistant about your ad accounts — spend, ROAS and campaigns across Google Ads, Meta, TikTok and 20+ more platforms — and make changes only after you approve them.
Remote MCP server and plugin for Claude, ChatGPT, Cursor, Gemini CLI and other AI assistants.
Quick start
You need an Orphex account with at least one workspace. Every client signs you in with that account on the Orphex sign-in page, in your browser; there is no API key to paste.
Try this first:
Using Orphex, how did our spend and ROAS change last week compared with the week before, platform by platform, and what drove the biggest change?
What you can do
Ask in your own words; the plugin's skills pick the prepared method, report template or live read that fits the question.
Platforms: Google Ads, Meta, TikTok, LinkedIn, Pinterest, Snapchat, X Ads, Apple Ads, Google Analytics 4, Shopify, AppsFlyer, Adjust and more than a dozen other ad, analytics and commerce platforms. Which of them a workspace reaches depends on what it has connected in Orphex.
Manual MCP setup
Orphex is a remote MCP server. In an MCP client that takes a server URL, add this one as a Streamable HTTP server and sign in with your Orphex account when the client asks:
Sign-in uses OAuth in your browser, never an API key. Orphex admits a known list of clients and does not accept dynamic client registration, so if your client cannot complete sign-in, write to [email protected].
Claude Code
Add this repository as a plugin marketplace, then install the plugin from it:
The first Orphex call opens your browser on the Orphex sign-in page. Claude Code receives
the sign-in on port 33418 of your machine, so that port has to be free while you sign
in. To pick up a newer release later, run /plugin marketplace update orphex.
Orphex events in your session (optional). The same marketplace carries a second
plugin, orphex-channel: a Claude Code channel that pushes what happens in your
workspaces (a change waiting for your approval, an ad disapproved, a budget running out)
into the session you have open. Install it with /plugin install orphex-channel@orphex and
see channel/README.md for how to turn it on; it needs Node.js 20 or later.
Claude apps
In claude.ai or Claude Desktop, connect Orphex by following the Claude setup steps at
https://mcp.orphex.co/support.
To upload the plugin to a Claude app, use the orphex-vX.Y.Z.zip attached to the
latest release.
Cursor
Add Orphex to Cursor
opens Cursor and offers to install the Orphex connection; the first call opens your browser
on the Orphex sign-in page. The same link as a Cursor deeplink is
cursor://anysphere.cursor-deeplink/mcp/install?name=orphex&config=eyJ1cmwiOiJodHRwczovL2FwaS5vcnBoZXguY28vYXBpL2ludGVyYWN0aW9uLWdhdGV3YXkvdjEvbWNwIn0=.
This repository is also laid out as a Cursor plugin, with its manifest in
.cursor-plugin/plugin.json, so Cursor reads the four skills from it as well.
Gemini CLI
Install this repository as a Gemini CLI extension, restart Gemini CLI, and sign in from inside it:
The extension names Orphex's public sign-in client itself, so your settings.json needs no
edit. The browser opens on the Orphex sign-in page, and Gemini CLI receives the sign-in on
port 7777 of your machine, so that port has to be free while you sign in. The extension loads
the four skills too.
Codex and ChatGPT
OpenAI's plugin directory is shared by ChatGPT and Codex. Once Orphex is listed there,
install it in Codex CLI by running codex and then /plugins, and in ChatGPT from the same
directory. The OpenAI build of this plugin is on this repository's openai branch.
What the plugin contains
Orphex is an advertising-analytics platform. This plugin teaches Claude how to work with an Orphex connection: how to find the right capability before calling it, how to address a workspace, how to read the metric vocabulary before the numbers, and which prepared method answers a question about spend, ROAS, conversions, pacing or anomalies across the ad, analytics and commerce platforms a workspace has connected. It also carries the approval-first flow for making a change you ask for, such as a budget, a bid or a campaign's status: Claude shows you a preview, waits for your approval, and only then makes the change.
- Four skills, written as plain Markdown instructions:
orphex-actions— makes a change you ask for, such as pausing or enabling a campaign, changing a budget or a bid, creating a campaign, ad set or ad, adding negative keywords, uploading media, changing Tag Manager, GA4, Singular or Insider catalog settings, or undoing an earlier change, through the approval-first flow described below.orphex-analyst— answers questions about what the numbers are, why they moved and what to do about it, and routes each one to the prepared method that fits.orphex-platform-reads— routes a question that names one advertising platform to the guide for that platform's own live reads.orphex-reporting— builds a report, such as a weekly or monthly performance report, an executive summary or an account review, by filling one of the report templates the workspace can use.
- One remote MCP server entry,
orphex.
It contains no hooks, commands, agents, scripts or executables, installs no packages, and runs no code on your machine.
These skills work through the Orphex connection. For portable marketing-analysis skills that work from exports you supply and use an Orphex connection when one is present, see Orphex Agent Skills.
Example prompts
Replace the names below with your own campaigns and workspaces. Each one names the skill that handles it.
- "Why did our ROAS on Meta drop last week compared with the week before?" — an
analysis, answered with the
orphex-analystmethod from read-only calls. - "Build the monthly performance report for September." — a report, built by
orphex-reporting, which starts from a prepared report template when the workspace has one. - "Which of our TikTok campaigns are active right now, and what budget is each one
on?" — a live read from the platform itself, routed by
orphex-platform-readsto the TikTok guide. - "Who paused my Google Ads campaigns this week?" — a read of what changed, combining
the platform's own change log with changes made through Orphex, routed by
orphex-analyst. - "Raise the daily budget of the 'Summer Sale' Meta campaign from 100 to 120, and show
me the change before you apply it." — a change, made by
orphex-actions: Claude shows the before-and-after preview and applies it only after you approve.
Security
How changes are made
Answering questions and building reports use read-only calls. A change happens only when
you ask for one, and the orphex-actions skill tells Claude to take the same steps every
time:
- Propose. Claude finds the change on the Orphex server, reads its procedure there and proposes it. For most changes the proposal changes nothing: the server answers with a preview of the item, its current value and the proposed one, and the result of its safety checks.
- Your decision. Claude shows you the preview and waits for your answer. The skill tells Claude to go ahead only on your approval and never to approve on your behalf.
- Change. Only after you approve does Claude confirm the change, with a one-time token bound to the preview you saw, so what runs is what you approved. The token expires about 15 minutes after the proposal.
- Outcome. Claude reads the result back from the change request's own status and tells you what happened, including any part that did not go through.
A few changes run in a single call with no separate preview; for those the skill tells Claude to describe what the call will do and to make it only after you say yes. Undoing a change is a new proposal that goes through the same preview and approval.
Changes go only through the Orphex endpoint below, and only where Orphex allows them: your Orphex plan has to include changes, the workspace has to have them switched on (they stay off until a workspace admin enables them), and your Orphex account needs an editor role in that workspace. Orphex's safety checks can refuse a change, for example a budget step larger than one change may take, and some workspaces require a recorded approval, sometimes from a second person, before a change can run. Where changes are not available, Claude tells you what the server says and changes nothing.
What it connects to
The plugin connects to exactly one endpoint, over HTTPS:
https://api.orphex.co/api/interaction-gateway/v1/mcp.
Sign-in uses OAuth through Orphex's Auth0 tenant at https://auth.orphex.co/: your
browser opens the Orphex sign-in page and you log in with your own Orphex account. The
plugin carries no password, API key or client secret.
What data is sent
The requests Claude makes on your behalf — which capability to run, the workspace, and arguments such as dates and metrics — go only to the Orphex endpoint above, and the results come back from it. Access is limited to the workspaces your Orphex account can already see. The plugin itself collects nothing, stores nothing and sends no telemetry.
Privacy
Orphex's privacy policy is at https://orphex.co/en/privacy-policy.
Requirements
An Orphex account with at least one workspace. Accounts and platform connections are managed at mcp.orphex.co.
Support
Questions and problems: [email protected].
About this repository
This repository is published from Orphex's own source by a release workflow, so changes are not made here and pull requests are not merged. Send problems and suggestions to the support address above.
License
Copyright 2026 Orphex. Licensed under the Apache License, Version 2.0; see LICENSE.
Source: README.md at commit 0131a6a
Tools
0Version history
2- v1.0.1LatestNov 7, 2026
- v1.0.0Oct 30, 2026
Score
85% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 daysRequired25 pts
- Auth declaredRequired10 pts
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts