Test deliverability from inside Cursor

Cursor is where a lot of email templates actually get written — React Email components, MJML files, raw HTML. The friction today is that once a template feels done you leave the IDE, open a deliverability tool, paste the compiled HTML, wait, screenshot, come back, edit, repeat. With Cursor's MCP support and our server the loop collapses into a single prompt: "test this template's deliverability and tell me what to fix". Here is the setup.

Cursor's MCP support in 2026

Cursor has had first-class MCP support since v0.45. Two surfaces use it:

• Agent mode — the autonomous "do this task" surface that can call tools, edit files and run commands. MCP tools show up as regular callable capabilities alongside Cursor's built-in file ops.
• Ask mode — the chat panel. MCP tools work here too, via @server-name mentions or by the model picking them implicitly.

Both surfaces read the same MCP config. You set it up once; it works everywhere.

Adding Inbox Check MCP to Cursor

Cursor supports two scopes for MCP: global (applies everywhere) and project (only when that repo is open). For deliverability I recommend project scope — it keeps the API key in your workspace's env, and different clients can use different keys.

Project scope: .cursor/mcp.json

Create .cursor/mcp.json at the repo root:

{
  "mcpServers": {
    "inbox-check": {
      "command": "npx",
      "args": ["-y", "ldm-inbox-check-mcp"],
      "env": {
        "INBOX_CHECK_API_KEY": "ic_live_REPLACE_ME"
      }
    }
  }
}

Add .cursor/mcp.json to your .gitignore if the API key is in there directly. Better yet, reference an env var:

{
  "mcpServers": {
    "inbox-check": {
      "command": "npx",
      "args": ["-y", "ldm-inbox-check-mcp"],
      "env": {
        "INBOX_CHECK_API_KEY": "${env:INBOX_CHECK_API_KEY}"
      }
    }
  }
}

Cursor expands ${env:VAR} from your shell environment — set INBOX_CHECK_API_KEY in your.zshrc or .bashrc and the config file stays safe to commit.

Global scope (alternative)

Open Cursor Settings → MCP → Add new MCP server. Fill in:

Name:    inbox-check
Command: npx
Args:    -y ldm-inbox-check-mcp
Env:     INBOX_CHECK_API_KEY=ic_live_...

Click save. Restart Cursor. The status dot next to the server name should turn green after a few seconds — that means the server started and successfully returned its tools/list.

Agent mode vs ask mode

Both modes can call the MCP tools; the difference is in how much autonomy you grant.

Ask mode

Chat-style. You ask, the agent proposes tool calls, you approve each one. Best for: exploring, reading a verdict, deciding what to fix.

Agent mode

The agent plans, executes and iterates. Best for: "edit the template until inbox rate is above 90%". With tool-use auto-approval enabled the agent can run a test, edit the HTML, re-run the test, and repeat — all without your intervention.

Example workflow: edit template, test, fix, retest

You're editing emails/welcome.tsx — a React Email component that compiles to HTML. Workflow:

1. Run your template build script in the terminal. The HTML lands in dist/welcome.html.
2. Open Cursor agent mode. Prompt: "read dist/welcome.html, run an Inbox Check placement test for it from hello@myco.com, wait for the verdict, and tell me the top three things to change".
3. Agent reads the file, calls start_test, polls get_test, summarises: "2/20 went to Outlook spam; SpamAssassin triggered HTML_IMAGE_ONLY_32; DKIM passes but DMARC alignment fails".
4. Ask: "fix the React Email component to have a 60/40 text-to-image ratio and a proper plain-text fallback". Cursor edits emails/welcome.tsx, rebuilds, reruns the test.
5. When the verdict is green enough, commit.

Prompts that work best

Cursor's agent is particularly good at multi-step reasoning when you write the goal, not the steps. Good prompts for this MCP:

• "Read templates/onboarding.html, run a placement test from welcome@myco.com, and open a file .claude/deliverability-report.md with the top-five issues sorted by severity."
• "Check SPF/DKIM/DMARC for every domain I send from (see config/senders.json) and tell me which ones need DMARC tightened."
• "Run a placement test for the HTML in out/marketing.html. If inbox rate is under 85%, iterate on the template until it's above 90%. Stop after three iterations."
• "Is our sending IP (203.0.113.45) on any blacklist? If so, write a remediation plan to docs/ip-delist.md."

Cursor vs Claude Desktop for this workflow

Both speak MCP; the tools are the same. The difference is context.

• Claude Desktop — best for ad-hoc debugging, pasting a template in, asking questions. No project awareness. No file editing.
• Cursor — best when the template lives in a repo and you want the agent to read it, test it, fix it, and commit. Project awareness is the differentiator.

For one-off checks use Claude Desktop. For iteration on an actual codebase, Cursor is the better fit.

Frequently asked questions