MCP

Follow these quick steps to integrate your AI client with Cursion MCP.

What it is

cursion-mcp is an MCP server that exposes Cursion platform actions as tools (Sites, Pages, Scans, Tests, Issues, Cases, Flows, and search).
It is a lightweight proxy layer: auth and data stay in the main Cursion API.

Hosted endpoint

Use the hosted Cursion MCP SSE endpoint:

https://mcp.cursion.dev/sse

You do not need to run the MCP server locally.

What you need

1. A Cursion account
2. Your Cursion API token from your account page

Authentication

Configure your MCP client to send:

Authorization: Token <your_api_token>

All tools are user-scoped, so actions are limited to resources your token can access.

Integration steps (any MCP client)

1. Add a new MCP server in your AI client
2. Set transport to SSE (or streamable HTTP if your client labels it that way)
3. Set server URL to https://mcp.cursion.dev/sse
4. Set auth header to Authorization: Token <your_api_token>
5. Save and reconnect, then verify tools are discovered

Available tools (high level)

cursion-mcp currently exposes these core tool groups:

• Sites: create_site, get_sites
• Pages: create_page, get_pages
• Scans: create_scan, get_scans
• Tests: create_test, get_tests
• Issues: create_issue, generate_issue, update_issue, get_issues
• Cases / CaseRuns: get_cases, create_caserun, get_caseruns
• Flows / FlowRuns: get_flows, create_flowrun, get_flowruns
• Processes / Search: get_processes, search

Example prompts

Once connected, try:

• "Find my sites"
• "Create a scan for site <site_id>"
• "Generate an issue from test <test_id>"
• "Search for checkout flow issues"

Built-in Cursion Agent

Inside the Cursion app Agent chat (/chat/:id), MCP integration is already wired in.
If you are using Agent in-app, you do not need to configure this manually.

Python Implementation with OpenAI

# import packages
from openai import OpenAI

# init OpenAI client
llm = OpenAI(api_key=YOUR_OPENAI_API_KEY)

# initial input string
input_string = (
    'BACKGROUND CONTEXT:\n'
    'You are a Software Quality Assurance Engineer.\n'
    'If necessary, call Cursion MCP tools to complete the task.\n'
)

# call llm with cursion-mcp as a tool
response = llm.responses.create(
    model='gpt-5-mini',
    input=input_string,
    tools=[{
        'type'              : 'mcp',
        'server_label'      : 'cursion-mcp',
        'server_url'        : f'{MCP_URL}/sse',
        'require_approval'  : 'never',
        'authorization'     : f'Token {CURSION_API_TOKEN}'
    }]
)

# log output
print(f'LLM response: {response.output_text}')