Use PageBolt as a native MCP server in Claude Desktop, Cursor, and Windsurf. Your AI agent can take screenshots and record videos natively.
# How to use PageBolt's MCP server in Claude Desktop, Cursor, and Windsurf
AI agents need web access. But running Chromium locally is expensive: 300MB per deployment, cold starts, memory spikes. Most agents can't do it.
MCP (Model Context Protocol) solves this. It's a standard that lets AI assistants call external tools natively—without installing anything.
PageBolt's MCP server gives your AI agent two superpowers:
1. **`screenshot`** — Take screenshots of any URL
2. **`record_video`** — Record browser workflows as MP4s with AI narration
No local Chromium. No infrastructure. One API call.
## Install PageBolt MCP in Claude Desktop
Claude Desktop auto-discovers MCP servers from a config file. Here's how:
**Step 1:** Create `~/.claude/mcp-servers.json`:
```json
{
"pagebolt": {
"command": "npx",
"args": ["@pagebolt/mcp", "--api-key", "YOUR_API_KEY"]
}
}
```
**Step 2:** Get your API key at [pagebolt.dev](https://pagebolt.dev)
**Step 3:** Restart Claude Desktop
That's it. Claude Desktop now has native access to PageBolt's screenshot and video tools.
## Using PageBolt in Claude Desktop
Open Claude Desktop and ask:
> "Take a screenshot of https://example.com and tell me what's on the page"
Claude will:
1. Call PageBolt's `screenshot` tool automatically
2. Download the image
3. Analyze it
4. Report back to you
No prompting needed. It just works.
## Example: AI Agent Workflow
Here's how an AI agent uses PageBolt to debug a deployment:
```
User: "Check if our landing page is live"
Agent (thinks): I need to screenshot the landing page URL
Agent: Calls PageBolt screenshot tool → gets image
Agent: Analyzes image → sees "404 Not Found"
Agent: Reports: "The landing page is returning 404. The deployment may have failed."
```
## Install PageBolt MCP in Cursor
Cursor uses the same MCP protocol. Same config file:
**Step 1:** Create `~/.cursor/mcp-servers.json`:
```json
{
"pagebolt": {
"command": "npx",
"args": ["@pagebolt/mcp", "--api-key", "YOUR_API_KEY"]
}
}
```
**Step 2:** Restart Cursor
Now Cursor's AI agent has native access to screenshot and video recording.
### Real Example: Cursor Agent Testing
```
Developer: "Use an agent to test the checkout flow"
Agent:
1. Screenshots the homepage
2. Sees "Add to Cart" button
3. Clicks it (via PageBolt record_video)
4. Records video of entire checkout flow
5. Saves MP4 to outputs/
Developer gets video proof that checkout works.
```
## Install PageBolt MCP in Windsurf
Windsurf is a Cursor fork. Same setup:
**Step 1:** Create `~/.windsurf/mcp-servers.json` (or use Cursor's config):
```json
{
"pagebolt": {
"command": "npx",
"args": ["@pagebolt/mcp", "--api-key", "YOUR_API_KEY"]
}
}
```
**Step 2:** Restart Windsurf
Windsurf's AI agent now has native screenshot and video tools.
## Why This Matters
Before MCP, AI agents were limited:
- They could analyze code, but not see what the code produced
- They could write tests, but not visually verify them
- They could deploy code, but not verify the deployment worked
With PageBolt MCP, agents can:
- Screenshot any URL instantly
- Record videos of workflows
- Verify their own work
- Debug without leaving the IDE
## Pricing
PageBolt MCP uses the same API:
- **Free tier:** 100 requests/month
- **Paid:** From $29/month (500 requests)
- **No setup fees, no infrastructure costs**
## Next Steps
1. Install PageBolt MCP in your IDE (Claude Desktop, Cursor, or Windsurf)
2. Ask your agent to screenshot a URL
3. Ask it to record a video of a workflow
4. Watch it work natively without any infrastructure setup
Your AI agent just got superpowers.
---
*PageBolt is the MCP server for web automation. Use it in Claude Desktop, Cursor, Windsurf, or any IDE that supports MCP.*