mcp-server-caniemail

MCP server giving an AI agent email client HTML/CSS compatibility data, backed by caniemail.com.

Email clients are not browsers. Outlook on Windows renders with Microsoft Word, support for anything modern is patchy, and the difference between "works with a workaround" and "nobody has ever tested this" decides whether an email ships broken.

Install

{
  "mcpServers": {
    "caniemail": {
      "command": "npx",
      "args": ["-y", "mcp-server-caniemail"]
    }
  }
}

Node 22+. Set CANIEMAIL_OFFLINE=1 to skip the network and use the bundled dataset.

Tools

• lint_email — give it drafted HTML and/or CSS plus a client list; returns only what breaks, with every source position the feature was used at, affected clients, documented workarounds, and a feature URL. Passing features are never returned.
• check_feature_support — per-client verdicts for one feature, for deciding how to build something rather than checking what you built.
• search_features — find feature slugs by keyword. Slugs are not guessable: "rounded corners" is css-border-radius, flexbox is css-display-flex.
• list_email_clients — the roster of 48 clients with display names. It is inlined into the other tools' descriptions, so this is rarely needed.

Four verdicts, not a boolean

| Verdict | Severity | Meaning | |---|---|---| | supported | — | Use it. | | unsupported | error | Will not render. Use a fallback. | | mitigated | warning | Works with a documented workaround — the note is the answer. | | untested | unknown | No data. Not evidence of support, or against it. |

Around a sixth of the matrix is untested. Results also carry last_test_date and a staleness note, because some entries have not been retested in five years.

Data is fetched from caniemail.com and cached for 24 hours; every result names which copy answered via data_source. The server revalidates as it runs rather than loading once at startup, so a session left open for days does not keep answering from the copy it fetched on the first morning. Startup itself touches no network — the handshake never waits on caniemail.com.

lint_email is sized for an agent's context. A finding carries only what its verdict decides — the severity, the clients affected, the notes — and everything that does not vary by verdict is stated once in a features legend keyed by slug: the title, the URL, the last test date, and positions, every place the markup uses the feature, as "line:col-line:col" strings with an occurrence_count. clients_affected is compressed against the clients you asked for ("*", "outlook.*", with client_count always exact), and the per-severity advice is a second legend, guidance, rather than a paragraph repeated on every finding.

Scope

Rendering only. Nothing about deliverability, SPF/DKIM/DMARC/BIMI, list management, or choosing between ESPs.

License

MIT. The caniemail dataset is a separate work — MIT, © 2019 Rémi Parmentier — fetched at runtime rather than redistributed here.

Source: https://github.com/shbernal/caniemail-ai-tooling