Help → Send to AI

Bring your Poggle context
into any AI.

Stop pasting the same notes into every chat. Three transports cover everything from a one-off ChatGPT question to a full-time Claude Code session — with audit, expiry, and one-click revoke baked in.

Open PoggleRead the API docs

Three transports

Pick the one that matches your tool.

Same data, three ways out. MCP for daily drivers, pull links for one-off chats, the terminal one-liner for engineers who never leave the shell.

Recommended
MCP — Claude Code & Cursor
Persistent connection, never expires, full read and write. The model discovers Poggle as a first-class tool and uses it natively.

When to pick it

You use Claude Code or Cursor regularly and want Poggle available in every session without re-pasting anything.

How

  1. Copy the config block below.
  2. Paste it into your client's MCP config file (Claude Code: ~/.claude/mcp.json, Cursor: Settings → MCP).
  3. Restart the client. Poggle's tools appear in the tool list.
{
  "mcpServers": {
    "poggle": {
      "command": "npx",
      "args": ["-y", "@poggle/mcp@latest"],
      "env": {
        "POGGLE_TOKEN": "pgl_live_…"
      }
    }
  }
}

Token from Settings → Connected apps.

Universal
Pull link — any chat tool
Zero setup. Works in Claude Web, ChatGPT, Gemini, or any model that can fetch a URL. One short line, one paste, the model reads the file itself.

When to pick it

One-off task, a tool that doesn't speak MCP, or you just want to share context with a teammate's chat without giving them a token.

How

  1. Open any note in Poggle and click “Send to AI.”
  2. Pick a TTL and copy the generated line.
  3. Paste it into your chat. The model fetches it on the next turn.
https://poggle.app/p/n/{token}.md
Terminal
Bash one-liner
Terminal-native. Pipes cleanly into the bash tool inside Claude Code, into a script, or into your own agent harness.

When to pick it

You live in a terminal, or you're wiring Poggle context into CI, a Make target, or a cron job.

How

  1. Generate a pull link the same way as above.
  2. Wrap it in curl and pipe to your tool of choice.
  3. Cap the byte count if you're worried about model context windows.
curl -s 'https://poggle.app/p/n/{token}.md.md' | head -c 50000

Update path

Updates land as proposals.

A pull link can be opened up to allow edits. They never write directly — every change queues for human approval.

How a write flows
  1. 1The AI POSTs an edit to your pull link (when edits are enabled).
  2. 2Poggle queues the change as a proposal — the note itself is untouched.
  3. 3Approve or reject from the AI Edits sidebar item. Approved proposals merge with full diff history.
  4. 4The AI sees the updated state on its next pull — same URL, fresh contents.
What gets recorded
Every read and every write hits the audit log — token fingerprint, user-agent, IP, byte count, and the resulting proposal id.
  • Read-only by default. “Allow edits” must be explicitly toggled on per token.
  • Approved proposals appear as a normal version on the note, attributed to the AI and the human who approved.
  • Rejected proposals stay searchable for 30 days so you can diff what the model wanted to do.

Expiry

How long does a link stay alive?

Four presets cover every realistic use. Idle tokens die on their own; nothing lasts beyond a day. For forever access, use MCP.

PresetTTLBest for
15 minHardOne-shot — “read this once.”
1 hourHardA single focused task.
SessionSliding 30 min, 24 h capSlidingActive agent work that should die when idle.
4 hoursHardLong task, no check-ins required.

Sliding window

Every read or write extends the window by 30 minutes. An idle agent’s token still expires on schedule — no orphaned live URLs.

Hard cap

Pull-link tokens never live beyond 24 hours, period. If you need persistent access, use MCP.

Security

Designed so you can revoke and forget.

A pull link is a scoped, hashed, expiring credential — not a permanent share.

Hash-only at rest
We store an Argon2 hash of every token. The raw value is shown exactly once, at creation. We can't recover it, and neither can a database leak.
Revoke any token instantly
Open Settings → Connected apps, click revoke. The next request returns 401, no propagation delay.
Every redemption logged
Time, user-agent, IP, byte count, status. Visible on the token detail page and exportable from the audit log.
Read-only by default
Edits require an explicit “Allow edits” opt-in, and even then they queue as proposals — a human approves before anything saves.
Try it: open any note and click Send to AI.
Thirty seconds from this page to a working pull link in your AI of choice. No credit card, no integration to write.
Open PoggleRead the API docs