Skip to main content
Give AI a budget, fill it cheaply, you pay and they deliver, unpack the week, enjoy dinner

handle-mcp

GitHub stars git clones PyPI version license MIT
macOS, Windows, Linux

Let your AI agent shop at nemlig.com, BilkaToGo and REMA 1000.
An MCP server for Claude, Grok or Codex. It reads the offers, plans the week on your budget and fills the cheapest cart. You pay.

Install → · handlemcp.dk

Example from the demo: the agent compares nemlig.com (693 kr), REMA 1000 (712 kr) and BilkaToGo (741 kr) and fills the cheapest cart; nothing is ordered. Prices are examples

It can never order or pay. You do. That code does not exist, and a test fails if it shows up (why).

It counts anonymous usage, and tells you so. No id, no personal data. Off with HANDLE_TELEMETRY=0. Exactly what is sent.

How it works

  1. Plan. Give your agent a budget. It reads this week's offers and the recipes, and plans the meals around them.
  2. Compare. It prices the whole list in every store: pack sizes, offers, delivery fee and minimum order.
  3. Fill. It fills the cheapest store's cart with your login. Only the cart. (At REMA 1000: your shopping list "handle-mcp".)
  4. Pay. You open the store, check the cart, pick a delivery or pickup time and pay.

Taste, budget and meal rules live in your own prompt or skill. The server finds and counts. Build your own skill.

Requirements

  • macOS, Linux or Windows (x64). Windows on ARM is not supported yet: a dependency (cryptography) has no ready-made build for it.
  • uv. It fetches Python 3.12+ if you do not have it.
  • An MCP client, Grok, Claude Code, Claude Desktop or Codex etc.
  • An account at nemlig.com or BilkaToGo with home delivery to your address, or at REMA 1000 (Køb & Hent or shopping in the store yourself).

Install

uvx handle-mcp@latest login nemlig
claude mcp add handle -- uvx handle-mcp@latest

@latest runs the newest release at each start. To choose when to update, use a version instead, e.g. handle-mcp@1.5.2 (more).

Then ask:

  • "We have 4,000 kr a month for food. Plan this week's meals." It asks who eats, what you like and which store, then plans on this week's offers and fills the cheapest cart after your yes.
  • "What is on offer this week? Any free gifts?"
  • "Which delivery times are cheapest tomorrow?"

Other stores, Claude Desktop, headless machines and AI agents (--from-env): setup. nemlig's offers and REMA 1000's search work with no login. No terminal? Join the waitlist.

Tools

Tool Writes? What
search_products no Price, price per kg/l, offers
search_recipes no The store's recipes (nemlig)
list_recipes no Featured recipes (BilkaToGo: the week's madplan), by tag (nemlig) or by category (BilkaToGo)
get_recipe no The products to buy for a recipe, for any number of persons
list_offers no This week's offers and free gifts
get_nutrition no Nutrition per 100 g/ml from the label
compare_basket no Prices a list in every store, fees included, cheapest first
get_cart no Cart lines, fees, total, in one store or every store
set_cart_quantities yes Sets quantities in the cart (REMA 1000: the list "handle-mcp"). Nothing else
add_recipe_to_cart yes Puts a recipe's products in the cart, after your yes. Nothing else
get_delivery_info no Chosen time, fees
list_delivery_slots no Free times, cheapest first (REMA 1000: Køb & Hent times)
get_orders / get_order no What you usually buy

Full arguments and fields: docs/tools.md.

Usage counts

Anonymous counts, so we can show the stores what handle-mcp sends them. On by default. Every event has version and date (day only):

Event When Fields
server_started once a day per machine os
first_tool_call the first tool call of the day (listing tools is not a call) tool
login_ok once a day per store store
cart_filled after 30 quiet minutes, only if every line went in; then nothing until the cart is empty again store, basket_total_dkk (nearest 50 kr), item_count
list_filled the same, for REMA 1000's shopping list store, basket_total_dkk, item_count

Never sent: names, addresses, phone, email, logins, product names, order ids or any id for you or your machine. Events go to Umami Cloud over HTTPS. Umami sees your IP to answer and to guess the country; it does not save it. Sends run in the background with a 2 second timeout, and a failed send never breaks a tool. Cart links carry utm_source=handlemcp.

Your AI tells you once: the first tool answer on a machine has a telemetry_notice. The only file is ~/.config/handle-mcp/telemetry.json: whether the notice was shown, the day each daily event was last sent, and which stores have a counted cart that is not empty yet.

Off:

HANDLE_TELEMETRY=0          # or DO_NOT_TRACK=1, or --no-telemetry

Your own Umami: HANDLE_TELEMETRY_URL and HANDLE_TELEMETRY_WEBSITE. Why: ADR 0005.

More

Setup · Skills · Security · Troubleshooting · Testing · Roadmap · Contributing · Credits

Unofficial. Uses the stores' private website APIs, so it can break any day. Not affiliated with nemlig.com, Salling Group or REMA 1000.

MIT

Free, and it stays free. If it saves you time or money: buy me a coffee

Metadata

Release files for handle-mcp 1.7.3

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for handle-mcp 1.7.3
File Size Uploaded Provenance Actions
handle_mcp-1.7.3.tar.gz 660.4 kB With provenance Details for handle_mcp-1.7.3.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for handle-mcp 1.7.3
File Interpreter ABI Platform Actions
handle_mcp-1.7.3-py3-none-any.whl Python 3 none any Details for handle_mcp-1.7.3-py3-none-any.whl

Total release size: 761.1 kB

Release files / handle_mcp-1.7.3.tar.gz

Download URL handle_mcp-1.7.3.tar.gz
Size 660.4 kB
Tags Source
SHA-256 checksum
How to use checksums
76c5e887a7f357dd88198a93158b940302bee6d2a4b40e2d789fc1d1532e3198
BLAKE2b-256 checksum
How to use checksums
54e15457d228682c83624d966e811bb1613a83950b3f747e62cf1af6ee063155
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 13, 2026.

Transparency log

Release files / handle_mcp-1.7.3-py3-none-any.whl

Download URL handle_mcp-1.7.3-py3-none-any.whl
Size 100.7 kB
Tags Python 3
SHA-256 checksum
How to use checksums
734de058a065a0a0d9512d440e43d3356a6b75fc3c70f72fd4a4ead533814137
BLAKE2b-256 checksum
How to use checksums
bf54f8fe8d51797036d3cf5a8e2973b723c1f70f8e0a9dd9c133b67c4f7233d7
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 13, 2026.

Transparency log

Security for handle-mcp 1.7.3

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for handle_mcp-1.7.3.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 13, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page