mcpField
G
GlockApps — Inbox Placement Testing

GlockApps — Inbox Placement Testing

Remote
glockapps.comBusiness & Marketing1Updated 4w ago

About GlockApps — Inbox Placement Testing

GlockApps MCP lets email teams run inbox-placement tests through an AI assistant and inspect results and suggested fixes across mailbox providers. It supplies real seed addresses but does not send campaigns.

Key features

  1. 01Create tests and retrieve seed addresses
  2. 02View results by mailbox provider
  3. 03Diagnose issues and prioritize fixes
  4. 04Compare test results
  5. 05Check projects, balance, and test status

Use cases

  1. 01Check whether a campaign reaches inboxes or spam
  2. 02Investigate filtering by specific providers
  3. 03Compare placement before and after a change

MCP support

The MCP server exposes GlockApps project, test, result, and diagnosis tools to AI clients. Clients can start tests and read reports; users send campaign email through their own platform.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://api.glockapps.com/mcp
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

Tools

Details

From the project README

GlockApps MCP

Run real inbox placement tests from your AI assistant. Ask it to start a test, and it hands you the seed addresses to send to; ask again once you have sent, and it tells you who put you in the inbox, who put you in spam, and what to fix first.

The seed network is live mailboxes at Gmail, Outlook, Yahoo, AOL, Zoho, corporate Exchange and 30+ other providers.

https://api.glockapps.com/mcp

This repository is documentation and the registry manifest. The server itself is hosted by GlockApps — there is nothing here to install or run.

What you need

  • A GlockApps account with API access on the plan. Your API key is on Account → API / MCP.
  • Test credits. Every test costs one, whether it is started here or in the GlockApps web app.

Quick connect

Claude.ai, Claude Desktop

Settings → Connectors → Add custom connector. Paste the address above and leave Client ID and Client Secret empty. Select Connect: a GlockApps page opens and asks for your API key once.

Claude Code
claude mcp add --transport http glockapps https://api.glockapps.com/mcp

Then /mcp in the session, and authenticate in the browser window it opens.

ChatGPT

Turn on developer mode (Settings → Security and login), then Plugins → + → Create app → Create MCP App. Paste the address, leave Authentication on OAuth, select Create, and paste your API key on the GlockApps page that opens. Developer mode is required while the connector is not yet in the ChatGPT directory.

Cursor, Windsurf, Continue and other clients

Clients that cannot do OAuth can send the API key directly. Copy a config from configs/ and put your key in it:

{
    "mcpServers": {
        "glockapps": {
            "url": "https://api.glockapps.com/mcp",
            "headers": { "Authorization": "Bearer YOUR_GLOCKAPPS_API_KEY" }
        }
    }
}

Full per-client instructions: docs/connect.md.

How a test works

The one thing to understand before you start: this connector cannot send your email. It gives you addresses to send to, and it reads what arrives.

  1. list_projects — resolves the project everything else needs.
  2. create_test — costs 1 credit. Returns the seed addresses and a marker.
  3. You send. Send the campaign you want to test, from the platform you are testing, to every seed address, with the marker included — either the X-API-Campaign-id header or the id: ... line in the body.
  4. wait_for_test — results are usually usable about 10 minutes after sending.
  5. get_results, then diagnose for what to fix, worst first.

A send without the marker never matches the test. Nothing arrives, the test never finishes, and the credit is still spent. If that happens, ask your assistant to delete the test: the credit comes back while no result has arrived.

Some providers accept mail slowly, so a test that has not "finished" after an hour is normal, not broken.

Things to ask it

  • "Start an inbox placement test on my main project."
  • "Has the test finished? What is the inbox rate?"
  • "Why did this campaign land in spam?"
  • "Compare this test with yesterday's. Did the fix work?"
  • "Which providers are filtering us, and what should I change first?"

Tools

Fifteen tools, thirteen of them read-only. Full reference with parameters: docs/tools.md.

Readlist_projects list_folders list_providers list_tests get_balance get_seed_list get_test_status wait_for_test get_results diagnose get_content_analysis compare_tests get_shared_link
Writecreate_test — costs 1 credit
Destructivedelete_test — removes the test and its results; refunds the credit only while no result has arrived

Security

  • Scopes. inbox_placement:read and inbox_placement:write. The connector cannot see or change your password, billing details or sending accounts.
  • It never sends email. Your campaign is sent by you, from your own platform.
  • Your key is not shared with the AI provider. In the OAuth flow it is exchanged on a GlockApps page for a token scoped to this connector; it stays on your GlockApps account.
  • Regenerating your key does not break an OAuth connection. The connector records which account it is linked to, not the key, and picks up the current one on each call. It does end any connection using the static-key method above.
  • Rate limits. Test creation is capped per hour and per day on top of your credit balance. get_balance reports the effective number.

Troubleshooting

"No test credits left" — top up in GlockApps, or run the test from the web app. get_balance tells you what is left before you start.

The test never finishes and shows nothing — almost always a send without the marker. Delete the test to get the credit back, start a fresh one, and follow its sending instructions exactly.

Some results missing after an hour — usually fine; slower providers take hours. Nothing at all within ten minutes of sending means the marker was missing.

"Shareable report links are unavailable" — not part of every subscription. Your assistant can still read per-mailbox results directly.

Connecting fails with "API key not recognised" — copy it again from Account → API / MCP; API access has to be enabled on your plan.

Support

Open an issue or email [email protected].

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.