---
title: Get Started with Pressable MCP
url: "https://pressable.com/knowledgebase/get-started-with-pressable-mcp/"
published: 2026-03-27
modified: 2026-08-17
author: Nox Dineen-Porter
---

## What Is Pressable MCP?

The **[Model Context Protocol (MCP)](https://modelcontextprotocol.io/docs/getting-started/intro)** is an open standard that allows AI assistants to connect to external services and take real actions on your behalf. The **Pressable MCP** connects your Pressable account to AI tools like [Claude](https://claude.ai/), [ChatGPT](https://chatgpt.com/), or [Gemini](https://gemini.google.com/app), letting you manage your WordPress sites through conversation rather than through the dashboard.

Instead of navigating menus to check PHP error logs, update a plugin, or create a staging site, you can simply ask your AI assistant to do it. For example:

- *“Update all my sites running PHP 8.2 to PHP 8.3.”*
- *“Show me plugin security alerts across all my sites.”*
- *“Create a staging site called client-preview in the Dallas datacenter.”*
- *“List the last 50 PHP fatal errors on the client-production site.”*

This is especially useful for agencies and developers managing many sites, since you can run bulk operations across your entire account without writing scripts or navigating the dashboard site by site.

AI agents and automation tools (such as [Claude Cowork](https://www.anthropic.com/product/claude-cowork)) can also be scheduled to perform ongoing tasks in the background, such as reviewing PHP and server logs at regular intervals and proactively alerting you to issues before they become large scale problems.

### What Can Pressable MCP Do?

At launch, the Pressable MCP supports the following categories of actions:

#### **Site Management**

- search and filter sites
- view site details and status
- create, clone, and delete sites
- enable and disable sites
- update PHP version
- toggle maintenance mode
- configure basic authentication
- manage PHP filesystem permissions
- toggle lightweight 404 handling
- toggle WordPress multisite support

#### **Performance and Monitoring**

- view Lighthouse performance reports (including Core Web Vitals)
- retrieve real-time metrics including response time, requests per second, bandwidth, PHP request rate, and MySQL CPU usage
- view storage and pageview statistics

#### **Backups**

- list and restore scheduled backups
- create, view, download, and delete on-demand backups (filesystem and database)

 Info:Each site supports up to 3 on-demand filesystem backups and 3 on-demand database backups.

If you need to create a new on-demand backup and have reached the limit, you can instruct your AI assistant to list the existing backups, identify the oldest one by date, delete it, and then create the new backup. The AI handles each step in sequence.

#### **Logs**

- retrieve PHP error logs (filterable by severity: Fatal error, Parse error, Warning, Deprecated, User)
- retrieve NGINX/webserver access logs (filterable by HTTP status code, request method, and IP address)
- view site-level activity logs
- view account-level activity logs

 
 Note on log retention: PHP error logs are available for 7 days, NGINX access logs for 7 days, site activity logs for 30 days, and account activity logs for 60 days.

If you need logs retained beyond these limits, you can schedule an AI agent to collect and export logs at regular intervals and save them to a file or external storage.

#### **Domains**

- list domains
- add and remove domains
- set the primary domain

#### **Plugins**

- list, install, activate, deactivate, update, and delete plugins
- check for plugins with known security vulnerabilities

#### **Themes**

- list, install, activate, update, and delete themes
- check for themes with known security vulnerabilities

#### **Collaborators**

- list collaborators, their permissions, and the sites they can access.
- add or remove collaborators from one or more sites.
- grant or revoke a specific permission across multiple sites without changing other permissions.
- add or remove collaborators across your entire account.
- optionally delete an associated WordPress user when removing a collaborator.
- report the progress and results of large changes, including any failures and their reasons.

Each collaborator and site is checked separately, so requests only affect sites you’re authorized to manage.

#### **WordPress**

- list WordPress users
- create a new WordPress admin user
- generate a One-Press login link for direct WP admin access

#### **SFTP**/SSH

- list SFTP/SSH accounts and view account details
- reset SFTP/SSH account password

## Requirements

- An active paid Pressable plan (MCP is available to all paying customers)
- Collaborators can also use MCP if their account owner has granted them the appropriate permissions
- **[Node.js](https://nodejs.org/)** is required on your own computer (not your Pressable server) for most AI clients, including Claude Desktop, Claude Code, and Gemini CLI. You can download it from [nodejs.org](https://nodejs.org/). Please note: Installing and configuring Node.js on your local machine is outside the scope of Pressable support.

## Step 1: Generate an Access Token

Most AI clients connect to Pressable MCP using a **bearer token**. Claude Code and ChatGPT (browser) use **OAuth** instead and do not require a token — see the Claude Code and ChatGPT sections below.

To generate a token:

1. Log in to [my.pressable.com](https://my.pressable.com/)
2. Navigate to **Tools → Pressable MCP**
3. Generate a new access token and copy it somewhere safe

![](https://i0.wp.com/pressable.com/wp-content/uploads/2026/03/mcp-auth-1.jpg?resize=1024%2C354&ssl=1)
 
 ![](https://i0.wp.com/pressable.com/wp-content/uploads/2026/03/mcp-auth-2.jpg?resize=1024%2C231&ssl=1)
 
  
 Note: **Treat your access token like a password.** Anyone with your token can manage your Pressable sites. Do not share it or commit it to version control.

## Step 2: Connect Your AI Client

### Claude Desktop — Bearer Token

Claude Desktop requires [Node.js](https://nodejs.org/) and a manual config file edit.

1. Generate an access token (see Step 1 above)
2. Install Node.js if not already installed: [nodejs.org](https://nodejs.org/)
3. Open your Claude Desktop config file: 
    - **macOS:** `~/Library/Application Support/Claude/claude_desktop_config.json`
    - **Windows:** `%APPDATA%\Claude\claude_desktop_config.json`
4. Add the following to the file, replacing `YOUR_TOKEN_HERE` with your actual token:

```
{
  "mcpServers": {
    "pressable": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.pressable.com",
        "--transport",
        "http-only",
        "--header",
        "Authorization:Bearer YOUR_TOKEN_HERE"
      ]
    }
  }
}

```

5. Save the file, then fully quit and reopen Claude Desktop
6. Ask Claude to do something to confirm it’s working (e.g., *“List my Pressable sites”*)

 
 Common error: If Claude Desktop shows a config error mentioning a missing `command` field, it means the config is using a `url`/`headers` format instead of the `command`/`args` format shown above. The `mcp-remote` bridge format above is required.

### Claude Code — OAuth

Claude Code connects via a terminal command. Node.js is required.

1. [Install Claude Code](https://code.claude.com/docs/en/quickstart) if not already installed: `npm install -g @anthropic-ai/claude-code`
2. Add the Pressable MCP server: `claude mcp add --transport http pressable <a href="https://mcp.pressable.com" rel="noreferrer noopener" target="_blank">https://mcp.pressable.com</a>`
3. Then start up claude in the terminal, and run this command: `/mcp`
4. You will see the new pressable MCP in the list. Click on it, and then click to authenticate. You will then be taken to the browser to authenticate.
5. Ask Claude to interact with your account to confirm it’s working

 
 Note: You can verify the connection in Claude Code by running `claude mcp list`. See the Claude Code section above.

### **Claude (Web) — OAuth**

No API key needed. Claude connects with OAuth using your Pressable login.

1. In Claude, open **Customize → Connectors** and click the **+**, then choose **Add custom connector**. *(On Team or Enterprise plans, an Owner or Primary Owner must first enable custom connectors under* ***Organization settings → Connectors****.)*
2. Enter the Pressable MCP server URL https://mcp.pressable.com and click **Add**. Leave the OAuth Client ID and Secret blank — they aren’t needed.
3. Sign in with your Pressable account and click **Authorize** to grant access.
4. To use it in a conversation, click the **+** in the message bar, choose **Connectors**, and toggle **Pressable** on.

### Codex — Bearer Token

1\. Generate an access token (see Step 1 above)

2\. Add an environment variable for your token:

Mac:

Open the terminal and run:

`launchctl setenv PRESSABLE_MCP_TOKEN "your-access-token"`

Windows:

Open the Command Line and run:

`setx PRESSABLE_MCP_TOKEN "your-access-token"`

Linux:

Create file: `~/.config/environment.d/pressable-mcp.conf`

The content of the file should be:

`PRESSABLE_MCP_TOKEN=your-access-token`

3\. In the Codex app, navigate to: settings / settings / MCP servers / + Add server

4\. Use the following settings, and click ‘Save’.

![](https://i0.wp.com/pressable.com/wp-content/uploads/2026/03/Screenshot-2026-06-10-at-4.01.16-PM.png?resize=1024%2C981&ssl=1)### Gemini CLI — Bearer Token

Gemini has no desktop app, so this is the only supported Gemini connection method. [Node.js](https://nodejs.org/) is required.

1. Generate an access token (see Step 1 above)
2. Install Node.js if not already installed: [nodejs.org](https://nodejs.org/)
3. Install Gemini CLI: `npm install -g @google/gemini-cli`
4. Add the Pressable MCP server: `gemini mcp add pressable https://mcp.pressable.com --transport sse`
5. Open `~/.gemini/settings.json` and confirm it contains the following (updating the token):

```
{
  "security": {
    "auth": {
      "selectedType": "oauth-personal"
    }
  },
  "mcpServers": {
    "pressable": {
      "url": "https://mcp.pressable.com",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN_HERE"
      }
    }
  }
}

```

6. Close and reopen Gemini CLI, then ask it to interact with your account to confirm

 
 Note: Gemini in the browser does not support MCP connections.

> 

### ChatGPT (Browser) — OAuth

ChatGPT in the browser uses OAuth, which means there is no config file or token required. This is the simplest setup of any supported client.

1. Log in to [chatgpt.com](https://chatgpt.com/) in your browser
2. Go to **Settings → Apps → Advanced** and enable **Developer Mode**
3. Click **Create App**
4. Enter the Pressable MCP server URL: `https://mcp.pressable.com`
5. Follow the OAuth prompt to authorize your Pressable account
6. Once connected, ask ChatGPT to interact with your account (e.g., *“List my Pressable sites”*)

 
 Note: MCP Servers is not available on all ChatGPT accounts yet. If you do not see it in Settings, the feature has not rolled out to your account.

The ChatGPT standalone app works once the app is set up via browser, but Developer Mode must be enabled through the browser; there is no toggle in the app’s settings.

### Grok — Bearer Token *(Advanced)*

1. Generate an access token (see Step 1 above)
2. Install Node.js: [nodejs.org](https://nodejs.org/)
3. Sign in to [grok.com](https://grok.com/)
4. Create a config file at `~/.grok/pressable-mcp-config.json`:

```
{
  "mcpServers": {
    "pressable": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.pressable.com",
        "--header",
        "Authorization: Bearer YOUR_TOKEN_HERE"
      ]
    }
  }
}

```

5. Start the local proxy (this must stay running while you use Grok with MCP): `npx @srbhptl39/mcp-superassistant-proxy@latest --config /full/path/to/pressable-mcp-config.json`
6. Install the **MCP SuperAssistant** Chrome extension, then refresh the Grok window and click the extension to activate the connection

 
 Note: Grok requires a local proxy process to run alongside the browser. This setup is more involved than other clients, and **we are not able to provide detailed support for connection issues specific to Grok.**

### Llama / Ollama — Bearer Token *(Advanced)*

1. Generate an access token (see Step 1 above)
2. Install [Ollama](https://ollama.com/download) and pull a model that supports tool use (Qwen 3.0 or later is recommended): `ollama pull qwen3.5`
3. Install Go: `brew install go` (macOS), then add it to your path: `echo 'export PATH=$PATH:$(go env GOPATH)/bin' >> ~/.zshrcsource ~/.zshrc`

4. Install MCPHost: `go install github.com/mark3labs/mcphost@latest`
5. Create a workspace directory and an `mcp.json` config file inside it containing:

```
{
  "mcpServers": {
    "pressable": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.pressable.com",
        "--header",
        "Authorization:Bearer YOUR_TOKEN_HERE"
      ]
    }
  }
}

```

6. Launch MCPHost with your model and config: `mcphost -m ollama:qwen3.5 --config /path/to/your/mcp.json`

 
 Note: This setup requires a locally running Ollama instance and the MCPHost tool. It is intended for users comfortable with command-line tools and local model configuration. **We are not able to provide detailed support for connection issues specific to Ollama/MCPHost.**

### Mistral — Bearer Token *(Advanced)*

The general approach requires a Mistral API key, your Pressable MCP token, Node.js, the `@mistralai/mistralai` package, and the `@modelcontextprotocol/sdk` package. You will need to write a small Node agent that passes requests from Mistral through to the Pressable MCP server.

 
 Note: Mistral does not have native MCP support. Connecting it requires building a custom Node.js agent. This is an advanced integration intended for developers comfortable building their own tooling. **We are not able to provide detailed support for connection issues specific to Mistral.**

## Unsupported Clients

The following AI tools were tested and confirmed **not to support MCP connections** at this time:

- **Gemini (browser):** The browser version of Gemini does not support MCP. Use Gemini CLI instead.
- **Cohere:** Cohere’s MCP integration requires changes to be made on the MCP server side, which is not something Pressable supports. There is no supported workaround at this time.

## Revoke All MCP Tokens

You can revoke all MCP credentials associated with your Pressable account from **Settings → MCP Authentication**. Click **Revoke All MCP Tokens** and confirm to immediately disconnect all MCP access, including tokens created directly and connected AI apps such as Claude or ChatGPT. This does not affect other account access. Revoked AI connections must be reconnected before they can be used again.

![](https://i0.wp.com/pressable.com/wp-content/uploads/2026/03/mcp-revoke-tokens.png?resize=1024%2C173&ssl=1)### When to Revoke All MCP Tokens

Use **Revoke All MCP Tokens** when you want to immediately cut off all MCP access to your Pressable account, such as:

- You suspect an MCP token or connected AI app may have been compromised.
- You want to disconnect previously authorized AI apps such as Claude or ChatGPT.
- You’ve created multiple MCP tokens and want to invalidate all of them at once.
- You’re no longer using MCP integrations and want to start with a clean slate.
- A team member or integration should no longer have access through MCP.

Revoking the tokens immediately stops MCP access, while other Pressable account access remains unaffected.

## Known Issues

As the Pressable MCP is currently in public beta, you may encounter some bugs. At present, we are aware of and working to resolve the following issues:

- **Plan name displays incorrectly:** In some clients, the plan name associated with a site may not return the correct value. This is a known data issue and will be addressed in a future update.
- **One-Press login link error:** Generating a One-Press login link via MCP currently produces a link that returns an error when used. The link is generated successfully but cannot be used to log in.

*For general Pressable support, visit [pressable.com/knowledgebase](https://pressable.com/knowledgebase/) or contact our support team.*