MCP server

The @pocketping/mcp server connects PocketPing to any Model Context Protocol client — Claude Desktop, Cursor, and a growing list of AI tools — so your assistant can triage your inbox and actually reply to visitors.

Unlike read-only support integrations, a PocketPing reply sent through the MCP fans out to the chat widget and every connected bridge (Telegram/Discord/Slack) — your assistant closes the loop, it doesn't just summarize.

Setup

First, create an API key: dashboard → Settings → API keys → Create key. Copy it (it's shown once).

Hosted (recommended)

The fastest path — no install. Add a custom connector pointing at the hosted server and authenticate with your key as a Bearer token.

• Connector URL: https://app.pocketping.io/api/mcp
• Auth: Authorization: Bearer ppk_…

For config-file clients (Claude Desktop, Cursor):

{
  "mcpServers": {
    "pocketping": {
      "url": "https://app.pocketping.io/api/mcp",
      "headers": { "Authorization": "Bearer ppk_your_key_here" }
    }
  }
}

The endpoint speaks Streamable HTTP (the current MCP transport; legacy SSE is disabled) and authenticates with your API key as a Bearer token — so it works in any client that lets you set a header or token: Cursor, Claude Desktop, the Claude API connector, ChatGPT, or the local option below. (Claude's web "Add connector" UI expects an OAuth flow rather than a Bearer header, so use one of those clients there.)

Local (npx)

Prefer to run it on your machine, or self-hosting PocketPing? Use the @pocketping/mcp package over stdio:

{
  "mcpServers": {
    "pocketping": {
      "command": "npx",
      "args": ["-y", "@pocketping/mcp"],
      "env": {
        "POCKETPING_API_KEY": "ppk_your_key_here",
        "POCKETPING_API_URL": "https://app.pocketping.io"
      }
    }
  }
}

Self-hosting? Point POCKETPING_API_URL at your own instance.

Tools

Tool	Read/write	What it does
list_projects	read	List your projects
list_sessions	read	List conversations — filter by projectId, status, unanswered, or a search q
get_conversation	read	Full transcript + visitor details for one session
get_stats	read	Mini support stats — conversations, response rate, median first reply, CSAT% (projectId?, period 7d/30d)
send_reply	write	Send a real reply to a visitor (widget + all bridges)

send_reply is the only write tool and is marked non-read-only, so well-behaved clients ask you to confirm before it sends.

Prompts

Prompt	What it does
triage_unanswered	Summarize chats awaiting a reply and propose drafts
draft_reply	Draft (but not send) a reply for a specific conversation

Example

You: Summarize my unanswered PocketPing chats and draft a reply to each.

The assistant calls list_sessions(unanswered=true), then get_conversation on each, summarizes, and proposes drafts. You approve one, and it calls send_reply — the visitor sees it in the widget, and it lands in your Telegram/Slack thread too.

Security

• Keys are organization-scoped, hashed at rest, and revocable at any time from the dashboard (Settings → API keys).
• The server talks to the authenticated /api/v1 management API over HTTPS.
• Rate-limited to 600 requests/minute per key.

Behind the scenes

Both modes expose the same tools. The hosted endpoint (/api/mcp) runs in-process; the npx package is a thin stdio wrapper over the same /api/v1 management API. So anything the MCP can do, your own scripts and agents can do too with the same API key.