Pp Instacart

Purpose

Printing Press CLI for Instacart. Natural-language Instacart CLI that talks directly to the web GraphQL API. Add items to your cart, search products, and manage carts across retailers without browser automation. Also caches your purchase history locally so 'add' resolves items you have bought before instead of guessing from live search. Trigger phrases: 'install instacart', 'use instacart', 'run instacart', 'add X to my Safeway cart', 'what did I buy last time', 'order the usual', 'add my regulars to Costco', 'backfill my instacart history', 'sync my instacart orders', 'download my order history', 'save my instacart history locally'.

Description

Instacart - Printing Press CLI

Prerequisites: Install the CLI

This skill drives the instacart-pp-cli binary. You must verify the CLI is installed before invoking any command from this skill. If it is missing, install it first:

  1. Install via the Printing Press installer: 
    npx -y @mvanhorn/printing-press install instacart --cli-only
  2. Verify: instacart-pp-cli --version
  3. Ensure $GOPATH/bin (or $HOME/go/bin) is on $PATH.

If the npx install fails (no Node, offline, etc.), fall back to a direct Go install (requires Go 1.26.3 or newer):

go install github.com/mvanhorn/printing-press-library/library/commerce/instacart/cmd/instacart-pp-cli@latest

If --version reports "command not found" after install, the install step did not put the binary on $PATH. Do not proceed with skill commands until verification succeeds.

When to Use This CLI

Reach for this when a user wants:

  • Add a product to an Instacart cart by natural language ("add lemon sorbet to QFC")
  • Add something they have bought before ("add my usual milk to Safeway")
  • Show, search, or compare their active carts across retailers
  • List or search their own Instacart order history
  • Run an Instacart flow from a script, cron job, or agent loop

Do not reach for this if the user wants to actually check out. This CLI adds items to your cart; you still complete checkout in the Instacart app or web UI.

Unique Capabilities

History-first add

add checks your local purchase history FIRST and, when a confident match exists at the target retailer, skips the three-call live GraphQL chain entirely. Drops the cost of "add the lemon sorbet pops I usually get" from ~1.2s to ~200ms AND makes it resolve to the right SKU (the one you actually buy) instead of whatever live search ranks highest today.

Confidence rules:

  • FTS5 match in your local purchased_items at that retailer
  • Purchased within the last 365 days
  • Was in stock on the last purchase

Falls through to today's live-search behavior when any condition fails. Pass --no-history to force live search.

Every successful add (history-resolved or live-resolved) writes back to purchased_items so the signal gets warmer without a full re-sync.

One-command history backfill

Typing "backfill my instacart orders" (or similar, see Argument Parsing) kicks off a Chrome-MCP-driven flow that walks the user's logged-in Instacart tab, extracts their order history into JSONL, and imports it into the local DB. After backfill, add resolves from real purchase history instead of live-search guesses.

Primary path: Chrome MCP. Fallback: paste three JS files into DevTools by hand.

Full walkthrough below under "Backfill Flow". Reference docs with more detail:

history list / history search / history stats inspect whatever has been loaded.

Instacart does not expose a clean order-history GraphQL op, so the legacy history sync command cannot work. See docs/solutions/best-practices/instacart-orders-no-clean-graphql-op.md for why.

Natural-language add

Resolves a product from free-text via Instacart's own three-call GraphQL chain (ShopCollectionScoped -> Autosuggestions -> Items) and fires UpdateCartItemsMutation. No browser automation.

When Instacart rejects a candidate with notFoundBasketProduct (autosuggest occasionally surfaces a product that is not addable at your active cart's shop), add automatically retries up to 3 ranked candidates before giving up. In --json output a successful retry sets retry_count > 0 and includes an attempts array listing the rejected item ids. When history-first resolution hits the same error, add falls through to live search and reports resolved_via: "history->live".

Multi-retailer carts

carts list shows every active cart across retailers at once. Useful for agents that need to know where items live before adding to the right one.

Command Reference

Authentication:

  • instacart auth login - extract session cookies from Chrome
  • instacart auth status - show current session state
  • instacart auth logout - clear saved cookies
  • instacart auth paste - paste cookie JSON manually (fallback for newer macOS Chrome)
  • instacart auth import-file <path> - load cookies from a browser-use export JSON

Cart operations:

  • instacart add <retailer> <query...> - add a product by natural language
  • instacart add <retailer> <query...> --no-history - skip the history-first resolver
  • instacart add --item-id <id> <retailer> - add by exact Instacart item id
  • instacart cart show <retailer> - show current cart contents at a retailer
  • instacart cart remove <item-id> <retailer> - remove an item from a cart
  • instacart carts list - list every active cart across retailers

Discovery:

  • instacart search <query> --store <retailer> - search products at a retailer
  • instacart retailers list - list retailers available at your address
  • instacart retailers show <slug> - cache one retailer locally

Purchase history:

  • instacart history import <path> - load a JSONL order dump into the local DB (the working path)
  • instacart history import - --json - read from stdin, JSON output for agent pipelines
  • instacart history import <path> --dry-run - preview counts without writing
  • instacart history list - top purchased items by count + recency
  • instacart history list --store <retailer> --limit 20 - filter + paginate
  • instacart history search <query> - FTS search your purchase history
  • instacart history search <query> --store <retailer> - scoped FTS search
  • instacart history stats - counts + per-retailer state

Maintenance:

  • instacart doctor - health check: config, store, ops, history, session, live ping
  • instacart capture - refresh the GraphQL operation hash cache
  • instacart capture --remote - merge fresh hashes from the community registry
  • instacart ops list - show the operation-hash cache state

Recipes

First-time setup

instacart auth login                # extract cookies from Chrome
instacart doctor                    # verify auth + live ping
instacart capture                   # seed built-in op hashes

Then backfill history (optional but recommended; unlocks history-first add):

Tell the agent: "backfill my instacart orders"

The skill drives the rest. See the "Backfill Flow" section below.

Add something you buy all the time

instacart add safeway "oat milk"    # resolves via local history if you have bought it before

Look for via history in the output. If you see via live, the FTS match did not pass the confidence check; check instacart history search "oat milk" --store safeway to see what is actually in your history.

Force a fresh live search

instacart add safeway "oat milk" --no-history --dry-run --json

--dry-run --json is useful when debugging - the output includes resolved_via so you can see which path would have fired.

Daily top-up from recent history

instacart history list --store safeway --limit 20 --json | jq -r '.[].name' \
  | while read item; do instacart add safeway "$item" --yes --json; done

Auth Setup

Requires a logged-in Instacart session in Chrome. The CLI extracts cookies via kooky (no credential handling on our side). If Chrome is locked or you are on a system kooky cannot read:

instacart auth paste         # paste the full cookie JSON manually
instacart auth import-file <path>

Session lives at ~/.config/instacart/session.json (0600).

Agent Mode

The CLI is agent-native by default. Pass --json on any command for machine-readable output. --dry-run previews add without firing the mutation and surfaces which resolver (history, live, or item-id) would have fired.

add JSON envelope fields worth knowing:

  • resolved_via: one of history, live, history->live (history pick was rejected, live retry succeeded), or item-id.
  • retry_count: how many candidates were rejected before the winner. 0 when the first pick landed.
  • attempts: present only when retry_count > 0, array of {item_id, name, error_type} for each rejected candidate.
  • On exhaustion (exit 5): JSON envelope with error, retailer, query, attempts, and a hint naming the concrete next step (search then add --item-id, or retry with --no-history).

Filtering output

--select accepts dotted paths to descend into nested responses; arrays traverse element-wise:

instacart-pp-cli <command> --agent --select id,name
instacart-pp-cli <command> --agent --select items.id,items.owner.name

Use this to narrow huge payloads to the fields you actually need — critical for deeply nested API responses.

Response envelope

Data-layer commands wrap output in {"meta": {...}, "results": <data>}. Parse .results for data and .meta.source to know whether it's live or local. The N results (live) summary is printed to stderr only when stdout is a TTY; piped/agent consumers see pure JSON on stdout.

Exit Codes

CodeMeaning
0Success
2Usage error
3Auth missing or rejected
4Resource not found
5API error / conflict
7Rate limited or transient network

Argument Parsing

Given a free-form natural-language request:

  1. Empty, help, or --help -> run instacart --help
  2. Starts with install -> CLI install; ends with mcp -> MCP install
  3. Matches a backfill intent -> run the Backfill Flow below. Trigger phrases include: "backfill my orders", "backfill my history", "sync my instacart history", "sync my instacart orders", "download my order history", "save my instacart history locally", "pull in my past orders", "import my recent orders".
  4. Anything else -> map to the best subcommand and run with --json when invoked from an agent

Backfill Flow

Drive this when the user hits a backfill intent. Read docs/backfill-walkthrough.md via WebFetch for the full procedure; summary below.

Setup check:

  1. Confirm instacart-pp-cli is on PATH. If not, install: go install github.com/mvanhorn/printing-press-library/library/commerce/instacart/cmd/instacart-pp-cli@latest.
  2. Probe mcp__claude-in-chrome__tabs_context_mcp. If the tool is unavailable, route to the DevTools fallback: fetch docs/backfill-devtools-fallback.md and walk the user through it. Stop.
  3. Run instacart-pp-cli history stats --agent. If orders > 0, this is a top-up run; the resume state will skip already-dumped orders automatically.

Chrome MCP loop:

  1. WebFetch the three JS files (cache each in the current session):
    • https://raw.githubusercontent.com/mvanhorn/printing-press-library/main/library/commerce/instacart/docs/dumper.js
    • https://raw.githubusercontent.com/mvanhorn/printing-press-library/main/library/commerce/instacart/docs/extract-one.js
    • https://raw.githubusercontent.com/mvanhorn/printing-press-library/main/library/commerce/instacart/docs/export-jsonl.js
  2. Open or reuse a tab at https://www.instacart.com/store/account/orders. If the dumper returns profile_picker: true, ask the user to pick a profile in the tab, then re-run.
  3. Inject dumper.js via mcp__claude-in-chrome__javascript_tool. Read back total_ids and pending_extract.
  4. For each order ID in the pending set, navigate to /store/orders/<id> then inject extract-one.js. Report progress to the user every 10 orders.
  5. When pending reaches 0, inject export-jsonl.js. It downloads instacart-orders.jsonl to the user's default Downloads folder.
  6. Run instacart-pp-cli history import ~/Downloads/instacart-orders.jsonl --agent in a Bash tool. Show the summary JSON to the user.
  7. Verify: instacart-pp-cli history stats --agent. Offer a follow-up sanity check: instacart-pp-cli add <retailer> "<something they've bought>" --dry-run --json and flag resolved_via: "history" when it appears.

Error surfaces worth translating for the user:

  • Extractor cache_key_missing on every order -> Instacart rotated their web bundle. Report the observed cache keys and point at the rotation-recovery section of the walkthrough doc.
  • Dumper reports fewer IDs than the user expected -> probably on a multi-profile account; ensure the selected profile is the one with purchase history.
  • history import shows 0 orders imported -> the JSONL is empty (only skip records). Re-run the extractor loop with fresh tabs.

Direct Use

  1. Check installed: which instacart-pp-cli
  2. Check auth: instacart doctor
  3. Capture GraphQL hashes: instacart capture
  4. (Optional but recommended) Backfill history — run the Backfill Flow above. Unlocks history-first add resolution.
  5. Run your command with --json if invoked from an agent

Source

Codex

Capabilities

  • See original source for capabilities.

Inputs

Inputs depend on the skill's trigger and arguments. See the source SKILL.md.

Outputs

Outputs depend on the skill. Typical: files written, reports generated, agent actions performed.

When To Use

When the user invokes /pp-instacart or describes a task the skill's description matches.

Dependencies

See the source skill's references and scripts folders.

Related Systems

  • Claude (if synced from ~/.claude/skills/pp-instacart)
  • HQ Project — landing page Skills section
  • MD Project (md.sgnk.ai) — Skills/Pp Instacart/
  • Obsidian Vault — Skills/Pp Instacart/

Examples

See workflow.md.

Source: ~/.codex/skills/pp-instacart/SKILL.md Category: Printing Press