The Depra AI MCP server lets an AI assistant read your Depra AI data and answer questions about it. It is read-only, has 24 tools and is included on every plan. Use it when you want to ask about your visibility inside Claude Code, Cursor or another AI app.
The server address is https://api.depra.ai/mcp. You sign in with a personal access token from the API Keys page, which sits in the Depra AI app's sidebar under Company.
What is MCP?
MCP is short for Model Context Protocol. It is an open-source standard that lets AI applications connect to outside systems, such as data sources and tools (Model Context Protocol, checked 7 Oct 2026). An MCP server offers a set of tools. An AI app that supports MCP calls those tools when your question needs them. With the Depra AI MCP server connected, your assistant can look up your own numbers.
What does the server give an AI assistant?
It gives the assistant 24 tools over your own brand data. Each tool returns one kind of stored data.
| Area | What the assistant can read |
|---|---|
| Projects and setup | Your projects, brand profile, tracked brands, topics, tags, prompts and the AI models your plan tracks |
| Visibility | Visibility, share of voice, sentiment and position for each tracked brand, and your Insights |
| Answers | Stored AI answers in full, with the brands named, the citations and the web searches ChatGPT ran |
| Sources | Cited domains and pages, the sources that cite competitors and not you, and stored page text |
| Shopping | The products ChatGPT showed, by brand, merchant, prompt and product |
| Perception and facts | The qualities answers tie to each brand, and your facts with supported and contradicted claims |
| Actions | The recommended actions with their evidence, steps and brief |
| Changes | What changed between the last two scan days |
Is the server read-only?
Yes. The server reads a project's stored data. It cannot change a project, start a scan or edit a website.
- No writes: no tool adds, edits or deletes anything in Depra AI.
- No live fetching: the tool that returns page text reads the stored copy and never fetches the live page.
- No scans: answers are collected on the normal schedule, not when the assistant asks.
Depra AI has no general REST API. The MCP server is the only programmatic access, which means the only way for another program to read your data.
Which apps can connect?
The API Keys page in the app has a setup tab for each of five apps, and an Other tab for the rest. Two apps cannot connect yet.
| App | Connects | How |
|---|---|---|
| Claude Code | Yes | As a remote HTTP server, with the token in a header |
| Cursor | Yes | As a remote HTTP server, with the token in a header |
| VS Code | Yes | As a remote HTTP server, with the token in a header |
| Claude Desktop | Yes | Through mcp-remote, an open-source bridge |
| Codex | Yes | As a remote HTTP server, with the token in a header |
| Other MCP apps | Steps on the Other tab | As a remote MCP server over Streamable HTTP, with the token in a header |
| claude.ai on the web | Not yet | Cannot connect yet |
| ChatGPT | Not yet | Cannot connect yet |
The pricing page also lists "other MCP apps" and does not name them. For those, the Other tab says to add a remote MCP server with the server address, pick Streamable HTTP if the app asks for a transport, and add the token as an Authorization header.
What does it cost?
Nothing extra. The MCP server is included on every plan, and it works while your trial or subscription is active. Prices are on plans and limits.
How does sign-in work?
You create a personal access token on the API Keys page, in the Depra AI app's sidebar under Company. A personal access token is a secret key that lets an app sign in to the server. Your AI app sends it in an Authorization: Bearer header.
- What a token reads: a token can read only the projects its creator can access.
- Expiry and Revoke: each token has an expiry date and a Revoke link. Revoke all my tokens revokes all of your own tokens at once.
- Members: owners and admins can switch off Members can create tokens and connect apps. When it is off, members cannot create tokens and their tokens stop working. Owners and admins keep access.
Warning: Treat the token like a password. Keep it secret and do not commit it to git, which means never save it in a code repository. Do not paste it into a chat, a screenshot or a file that other people can open.
How to create, revoke and limit tokens is on connect your AI app.
What should I know about data safety?
Some of what the tools return was not written by you or by Depra AI: AI answers, and text copied from websites. Such text can contain a sentence that is written to look like an instruction to an AI assistant.
The server labels this text. Fields named untrusted_* hold text copied from websites and AI answers, or drafted by a model from them. An assistant should treat them as data and never follow instructions inside them.
- The server cannot act: it is read-only, so no text can make it change your project.
- Your app may have other tools: if your assistant can also edit files or send messages, review what it does after it reads web text.
- Results go to your AI app: what the tools return is passed to the app you connected. Check that app's own data terms.