Penlog for agents & developers

Penlog is a handwriting app for iPad. You write by hand with Apple Pencil; Penlog OCRs each page into structured structure_json — typed lines with sections, notation, and task status. That structure is what makes your handwriting addressable: any MCP-compatible agent can read your pages and write typed blocks and tasks back alongside your ink.

These docs are for people building on top of that — connecting an existing agent (Claude, ChatGPT, a CLI client) or authoring their own integration.

The mental model

  • You write ink. Agents write typed blocks. The two never collide — ink is yours alone, so there are no merge conflicts by design.
  • Every page is a resource. A page is identified by its journal and date, and exposes its OCR'd lines, tasks, and events.
  • Notation drives structure. The symbols you draw — a bullet, a circle, a dash — deterministically map to sections and status. Agents read the result, they don't have to guess.

Two ways to connect

  • One-click OAuth (recommended) — add Penlog as a connector in claude.ai, ChatGPT, or Claude Desktop and sign in with Apple at mcp.penlog.app. No tokens to copy.
  • Bearer token — for CLI and self-hosted MCP clients, mint a plg_live_ token and point your mcp.json at mcp.penlog.app.

Start here

Quickstart Add Penlog as a connector via one-click OAuth. MCP tool reference The six v1 tools, with input/output schemas and examples. Notation & structure_json How handwriting becomes structured, addressable lines. Data model Journals, pages, blocks, tasks, events, and seeds. Connector authoring The ConnectorService protocol and how Notion / Calendar plug in. Agent recipes Morning brief, weekly-review harvest, seed-from-agent. Remote MCP clients mcp.json + bearer token for CLI and self-hosted clients. Troubleshooting Connection, auth, Pro-gate, and empty-result issues, and how to fix them.