NAPSAC — Notion Agent Persistent Store for Accessible Context

Memory Backend

NAPSAC uses Notion exclusively. All reads and writes go through Claude's native Notion connector (Notion MCP tools).

Required Notion MCP tools:

• notion-search — Find pages by title
• notion-get-page / notion-fetch — Read page content
• notion-create-page / notion-create-pages — Create new pages
• notion-update-page — Update existing pages

No other infrastructure. No GitHub, no local server, no API tokens, no config files. If the user has Notion connected, NAPSAC works.

Page Structure

Memory is organized as Notion sub-pages under a root page:

[Memory Root Page]
├── _index (auto-managed, lists all files with links)
├── _conventions (naming/formatting rules — system page)
├── contacts/
│   └── contacts/key-people
├── context/
│   ├── context/general
│   └── context/preferences
├── profiles/
│   └── profiles/mynah-styles
└── projects/
    ├── projects/project-alpha
    └── projects/project-beta

Conventions:

• "Directories" are Notion pages whose title ends with / and contain child pages. They are organizational containers only.
• "Leaf pages" contain actual memory content. Their title is the full file path (e.g., context/preferences).
• The _index page is auto-generated on every memory_update.
• The _conventions page is created once during memory_init and never modified during normal operations. It is a system page.
• Page titles use the file path as the identifier for easy lookup.

Tool Implementations

memory_init

Purpose: Set up a new memory root in Notion.

Input:

• page (required) — Notion page URL or ID to use as the memory root

Behavior:

1. Validate that the Notion connector is available. If not, return: "Notion connector is not connected. Enable it in Settings → Connectors → Notion to use NAPSAC memory."
2. Fetch the provided page to confirm it exists and is accessible.
3. Create three child pages under the root:
   • _index — Empty index page (will be populated on first update)
   • _conventions — Naming and formatting rules (see _conventions content below)
   • context/general — Starter memory file with content:

     ## General Context
     
     This is the default memory file. Save general context and preferences here.
     
     *Created by NAPSAC on [date].*
     

4. Create the context/ directory page as a container.
5. Store the root page ID internally for subsequent calls.
6. Return confirmation with the root page URL.

_conventions page content:

# NAPSAC Memory Conventions

This page defines how memory is structured, named, and formatted.
Every AI tool reading this memory must follow these rules.

## Page Naming
- Leaf pages use full paths as titles: context/preferences, projects/alpha
- Directory pages use trailing slash: context/, projects/
- No file extensions (Notion pages don't have them)
- _index and _conventions are system pages -- never delete or rename

## Directory Taxonomy
- context/ -- Stable background: role, preferences, tooling
- projects/ -- Active and archived project memory
- profiles/ -- Style/format profiles (MYNAH, BINER)
- contacts/ -- People, accounts, org references
- New directories can be created freely

## Content Format
- Write as native Notion blocks, not raw markdown
- Headings, bullets, code blocks, quotes use native block types
- Pages must be readable in Notion UI without cleanup

## Index (_index)
- Regenerated after every memory write
- Lists all files with: path, description, tags, last updated
- Target: ~700 tokens for ~25 files
- If _index drifts, the next write auto-corrects it

## Profiles
- MYNAH: profiles/mynah-{context-id} (schema matches PACK)
- BINER: profiles/notion-design (schema matches PACK)
- Profiles are portable between PACK and NAPSAC

Error handling:

• If the page URL/ID is invalid or not found: return a clear error with the provided value.
• If Notion connector is unavailable: return the standard connector error.

memory_list

Purpose: Return a lightweight list of all memory files. This is the first call at every session start.

Input:

• tag (optional) — Filter by tag
• topic (optional) — Filter by topic
• dir (optional) — Filter by directory prefix (e.g., projects/)

Behavior:

1. Find the _index page under the memory root.
2. Read its content and return it.
3. If filters are provided, parse the index and return only matching entries.

Output format: The index content, which lists all files with:

• File path
• Description (first line of content or explicit description)
• Tags (if present in content)
• Last updated timestamp
• Link to the Notion sub-page

Target: ~700 tokens for ~25 files.

Error handling:

• If root page is not set: "Memory root not configured. Run memory_init with a Notion page URL to get started."
• If _index not found: "No index found. Run memory_init to set up your memory root."

memory_get

Purpose: Read a specific memory file by path.

Input:

• path (optional) — File path (e.g., context/preferences). If omitted, concatenates all files and returns them (v1 compat).
• all (optional, boolean) — If true, concatenate and return all files.

Behavior:

1. If path is provided: a. Search for a child page under the root whose title matches the path. b. Path resolution: If no exact match, attempt fuzzy resolution:
   • If path has no /, search for context/{path}, projects/{path}, etc. If exactly one match is found, use it. If multiple matches, return an error listing the ambiguous options. c. Fetch the page content. d. Convert Notion blocks to clean readable text. e. Return the content.
2. If path is omitted or all is true: a. Read the _index page. b. Fetch all linked pages. c. Concatenate their content with section headers. d. Return the combined content.

Output: