Model Context Protocol

Many patent professionals already work with an AI assistant such as Claude or ChatGPT every day. The Model Context Protocol (MCP) is an open standard that lets such an assistant read data from and interact with other applications. A server exposes a set of tools, and the assistant discovers and calls them on your behalf. For Patalyze that means your assistant can search patents, chart claims and build up a research without you ever opening our app.

We offer two MCP servers:

If you only want access to our corpus of patents, simply install Patents (by Patalyze). It is read-only, and no data is stored or logged on our end: you make a request and receive patent data back.

If you want your assistant to persistently store, organize, enrich and visualize your patent research, use the Patalyze server. It lets an assistant do everything a research agent can do in the Patalyze app, with your own permissions. The assistant drives the agent loop while Patalyze provides the data and the tools.

Either server works with any MCP client. The Claude and ChatGPT sections below walk through the setup for each, and the rest of this page is the reference they build on.

Set up your client#

The setup is the same in every client: add a server URL and approve access in the browser. The simplest way to get started is the "Open in" button at the top of this page, which hands this page to Claude or ChatGPT so it can walk you through the steps. Claude Code and Codex can even add the servers themselves with the commands below, while Claude on the web and ChatGPT need you to paste the URL into their settings once.

Be aware that Claude and ChatGPT both call a remote MCP server a connector, so that word below always refers to a setting inside their apps. Patalyze has a Connectors section in its own settings that works in the opposite direction, attaching other MCP servers to your research agents.

Claude#

Connectors live on your Claude account, not on any single app. Add them once, in Claude on the web, in Claude Desktop or in Claude Code, and they are available everywhere you sign in, including Cowork and the mobile apps.

To use a connector in a conversation, open the "+" menu in the chat composer, then "Connectors", and toggle the server on or off for that chat.

Custom connectors work on every Claude plan, but the Free plan allows only one. In that case, add the Patalyze server: it includes corpus search, so it covers everything Patents (by Patalyze) does. Pick Patents (by Patalyze) instead if you specifically want a connector that cannot modify your research.

On Team and Enterprise plans, members cannot add custom connectors on their own. An Owner or Primary Owner adds each server for the organization first, under Organization settings > Connectors: choose "Add", then "Custom" > "Web", and paste the server URL. Each member then opens Customize > Connectors, finds the connector (it carries a "Custom" label) and clicks "Connect" to authenticate with their own Patalyze account.

Claude Code#

Connectors from Claude on the web are synced to Claude Code automatically. If you only use Claude Code, you can install both servers from the CLI instead:

claude mcp add --transport http patents https://data.patalyze.com/mcp
claude mcp add --transport http patalyze https://api.patalyze.com/mcp

ChatGPT#

You add the connectors in ChatGPT on the web (chatgpt.com). They live on your ChatGPT account, so once added they are also available in the mobile apps. Codex keeps its own configuration and is set up separately through its CLI.

ChatGPT reaches the servers through Developer mode, which registers a remote MCP server as a custom connector and exposes all of its tools. ChatGPT renamed connectors to apps in December 2025, so you will see both words in its settings; they mean the same thing. Be aware that Developer mode is available on the Plus, Pro, Business, Enterprise and Edu plans, but not on Free or Go.

Connectors are off by default in each new chat. To turn them on, click the "+" button in the chat composer, open "More" (shown as "Developer mode" on some plans) and toggle the servers on for that chat. A few things to know:

• Naming a server in your prompt helps ChatGPT pick the right tools, e.g. asking it to use Patents (by Patalyze) to find recent lidar patents from Bosch.
• ChatGPT asks you to confirm before running any Patalyze tool that modifies data. Patents (by Patalyze) is read-only and never triggers a confirmation.
• Memory switches off in these chats. That is Developer mode working as designed.

On Business plans, only workspace admins can use Developer mode. On Enterprise and Edu, an admin must first allow it under Workspace settings > Permissions & roles > Connected data. Authorized members can then switch it on in their own settings, and an admin can publish a vetted connector to the whole workspace.

Developer mode is one of two ways in. ChatGPT also reads deep research connectors, which it drives through two tools named search and fetch. Patents (by Patalyze) exposes both, so you can add it as a deep research connector and get patents cited with links back to patalyze.com. The Patalyze server has a wider tool set that deep research does not read, so reach it through Developer mode.

Codex#

Codex keeps its own MCP configuration, so connectors you added in ChatGPT do not carry over. Add the servers with the CLI:

codex mcp add patents --url https://data.patalyze.com/mcp
codex mcp add patalyze --url https://api.patalyze.com/mcp

The entries are written to ~/.codex/config.toml and apply to all your projects. The Codex IDE extension reads the same file, so the servers appear there too. To limit a server to one project, put its entry in that project's .codex/config.toml instead. Codex reads project-level config only in projects you have marked as trusted.

Other MCP clients#

Any client that supports remote MCP servers over HTTP with OAuth can connect. Add a server URL, https://data.patalyze.com/mcp for Patents (by Patalyze) or https://api.patalyze.com/mcp for Patalyze, and approve access in the browser. Many clients accept a JSON config like this:

{
  "mcpServers": {
    "patents": {
      "type": "http",
      "url": "https://data.patalyze.com/mcp"
    },
    "patalyze": {
      "type": "http",
      "url": "https://api.patalyze.com/mcp"
    }
  }
}

Patents (by Patalyze)#

Read-only access to the global patent corpus and the CPC/IPC classifications. Connect it and start by asking about a company's patent portfolio. No data is stored or logged on our end, and the server never touches your research, so there is nothing it can change. Even after you install the Patalyze server, Patents (by Patalyze) stays useful for quick lookups mid-research.

Access rights#

The server asks for a single scope when you connect:

Tools#

Three tools, each billed as one Data API request. Every plan includes monthly AI & Data usage, so see the pricing page for the limits.

Under the hood this is the Data API: the tools run the same search engine over the same global corpus, and each maps to one of its REST endpoints. The patent filter syntax is identical, so the endpoints reference doubles as the field and operator vocabulary for search.

Patalyze#

The full workbench. This server opens your research to an MCP client: patents, products, the mappings between them, attributes, tables, dashboards and notes. An assistant acts with your permissions and can read and edit exactly what you can, nothing more. This is the server that lets an assistant build out a research end to end.

Access rights#

The server asks for the scopes below when you connect. You can see every client you approved and revoke its access at any time under Settings > Connected apps.

Tools#

The tool list is short by design, because the workbench is one tool. execute runs a JavaScript script against your research, and everything in the data model is a plain function inside that script rather than a tool of its own. learn is the reference for it: call it with the topic sdk for every function with its arguments, or with no topic for the index of everything else it can teach, including the research skills your organization saved. The remaining tools are the things a script cannot do, such as driving a live browser, e-mailing a result or writing a memory.

Clients that support MCP Apps render what show returns as an interactive view, editable in place when you have write access. It comes with a small token tool that refreshes the short-lived token such a view holds; it is called by the view, never by you. notify is rate limited, and every recipient can turn these e-mails off under Settings > Notifications.

Usage costs#

With an external client, the model's tokens are billed by Claude, ChatGPT or whichever assistant you use, since the assistant drives the agent loop. On our side you only pay the service fees for what the tools actually do, priced exactly as in the app:

• Patent searches: €0.075 per search, on both servers.
• Product searches: €0.075 per product found.
• Web searches: €0.025 per search, €0.05 per deep search.
• Browser sessions: €0.15 per session of fifteen minutes.
• Uploaded files: €0.002 per page extracted from a document or image.

Every plan includes monthly AI & Data usage, and usage tells you what is left. See the pricing page for the limits and top-ups.