← Back to ChoppedFork
Beta

ChoppedFork AI integrations

Connect ChoppedFork to compatible AI assistants using the Model Context Protocol (MCP).

An authorized AI assistant can work with your recipes, pantry, shopping lists and meal plans while ChoppedFork continues to enforce your account permissions.

What can an AI assistant do?

Depending on the permissions available to the integration, an AI assistant can help you:

  • find and read your recipes;
  • see which households you belong to;
  • check what is available in your pantry;
  • suggest meals based on your pantry;
  • work with your shopping list;
  • read and update your meal plan;
  • create and edit recipes;
  • rate or favorite recipes;
  • perform other supported ChoppedFork actions.
You stay in control. Write and delete operations require the relevant permissions. You should review important or destructive actions requested through an AI assistant.

Connect Claude

Claude is currently the easiest tested way to connect to the hosted ChoppedFork MCP server.

✓ Tested with ChoppedFork
  1. Open Claude.
  2. Open Customize → Connectors.
  3. Choose to add a custom connector.
  4. Enter the ChoppedFork MCP server URL shown below.
  5. Enable authentication.
  6. Enter the OAuth Client ID shown below.
  7. Leave Client Secret empty.
  8. Connect and sign in using Google when prompted.

MCP server

https://mcp.choppedfork.com/mcp

OAuth Client ID

choppedfork-claude-public

Client Secret: leave this empty.

After authorization, Claude can use the ChoppedFork tools available to your account.

Try it

After connecting, try prompts such as:

Which ChoppedFork account am I connected to?
List my ChoppedFork households.
What recipes can I cook with what I have in my pantry?
Show me my shopping list.
Suggest dinners for this week based on my pantry.

Local AI with OpenWebUI

✓ Tested with OpenWebUI and a local Qwen model

ChoppedFork has also been tested with a self-hosted OpenWebUI installation using a local tool-capable language model.

OpenWebUI can connect to the same hosted ChoppedFork MCP endpoint: 

https://mcp.choppedfork.com/mcp

In OpenWebUI, add ChoppedFork as an external MCP tool server, configure OAuth authentication, connect your ChoppedFork account and enable the integration for the chat.

Advanced setup: OAuth callback addresses vary between self-hosted OpenWebUI installations. Public self-service OAuth client registration for arbitrary OpenWebUI installations is not yet documented by ChoppedFork. The integration itself has been successfully tested with a registered OpenWebUI OAuth client.

Use native function calling

The selected local model must support tool/function calling. In OpenWebUI, use Native function calling when supported by the model and inference backend.

Context window

MCP tools add their schemas to the model context. ChoppedFork exposes several tools, so a small context window may cause tool definitions to be truncated before they reach the model.

A context window of at least 16,384 tokens is recommended for the current ChoppedFork tool set when used together with OpenWebUI.

For an Ollama-compatible server, a configuration may look like: 

environment:
  - OLLAMA_NUM_CTX=16384

You can check for prompt truncation with: 

docker logs ollama 2>&1 \
  | grep -Ei 'ctx-size|n_ctx|truncating input prompt' \
  | tail -30

A message such as: 

truncating input prompt

indicates that the configured context window is too small for the prompt and tool definitions being sent to the model.

Personal API keys

ChoppedFork also supports personal API keys for advanced integrations.

API keys are separate from the hosted OAuth MCP connection. They can be useful for custom applications or integrations where you explicitly configure the credential yourself.

API keys can be created from your ChoppedFork Account page and can be restricted to specific write permissions.

Keep API keys secret. Treat an API key like a password. Do not publish it, include it in screenshots, commit it to source control, or share it with services you do not trust.

Permissions and security

ChoppedFork integrations are designed around scoped access. An integration does not automatically receive unrestricted access merely because it is connected.

Depending on the integration and authorization, permissions may include read access and separate permissions for actions such as:

  • changing pantry contents;
  • changing shopping lists;
  • changing meal plans;
  • creating or editing recipes;
  • deleting recipes.

Independent AI services may process information under their own privacy policies and terms after you authorize them to receive ChoppedFork data.

For more information, see the Privacy Policy and Terms of Service.

Troubleshooting

The integration says Connected, but the AI does not use tools

Confirm that the integration is enabled for the current chat and that the selected model supports native tool calling.

The model says a ChoppedFork tool does not exist

Check that the MCP integration is enabled and that the model receives the tool definitions. For local models, also check that the context window is large enough.

The model produces strange or unrelated output

Check the inference-server logs for prompt truncation. A context window that is too small can cut off tool schemas or other important prompt content.

Authentication is required again

OAuth sessions may expire or require reauthorization. Reconnect the ChoppedFork integration and complete the sign-in flow again.

Compatibility

ChoppedFork currently has verified MCP interoperability with:

  • Claude;
  • OpenWebUI with a local tool-capable model.

Other MCP-compatible clients may also work, but have not necessarily been tested by ChoppedFork.