Question 1

How do I disable n8n MCP Apps result cards and fix the 'Couldn't open n8n-mcp' widget errors?

Accepted Answer

Upgrade to n8n MCP 2.92.0 or later. To disable UI cards for the entire server, set `N8N_MCP_DISABLE_UI_APPS=true` before starting the process; only the exact value `true` is accepted. This applies to both stdio and HTTP deployments and overrides any per-request `uiAppsEnabled: true`. For per-request control, set `InstanceContext.uiAppsEnabled` to `false`. The server then stops advertising `_meta.ui`, `_meta['ui/resourceUri']`, `_meta['n8n-mcp/toolName']`, and `ui://n8n-mcp/*` resources, while all tools and text results remain available. Existing `resources/read` requests for `ui://` URIs continue to succeed so cached clients do not generate another error card. Start a new ChatGPT/MCP session if the client has cached the old tool-to-card mapping.

Question 2

How do I fix validate_node returning "Code cannot be empty" for pythonNative Python Code nodes in n8n-mcp?

Accepted Answer

The validator only checks `language === 'python'`, but n8n 2.x uses `pythonNative`, so it incorrectly reads the `jsCode` field instead of `pythonCode`. Fix `src/services/config-validator.ts` in `validateCode` by normalizing the language:

```ts
const language = config.language === 'pythonNative' ? 'python' : config.language;
const codeField = language === 'python' ? 'pythonCode' : 'jsCode';
```

Alternatively, use:

```ts
const codeField = ['python', 'pythonNative'].includes(config.language)
  ? 'pythonCode'
  : 'jsCode';
```

As a temporary workaround, normalize `pythonNative` to `python` before calling `validate_node`. Add a `validate_node` regression test using `language: "pythonNative"` with non-empty `pythonCode`.

Question 3

Why does $jmespath() return null in n8n expressions even when the node succeeds?

Accepted Answer

n8n intentionally swallows JMESPath parse errors inside `{{ }}` expressions, so an invalid query resolves to `null` while the node still reports success. In Filter/IF nodes, that `null` makes every item fail the condition.

Fix the JMESPath syntax:

- Wrap numeric literals in backticks:  
  `{{ $jmespath($json, "customers[?revenue > `100000`].name") }}`
- Use single quotes for string literals; double quotes are treated as field identifiers:  
  `{{ $jmespath($json, "customers[?country=='PL'].name") }}`
- Use JMESPath logical operators `&&` and `||`, not `and`/`or`:  
  `{{ $jmespath($json, "customers[?country=='PL' && active].name") }}`
- Use `==`, not a single `=`.
- Keep the argument order as object first, query second:  
  `$jmespath($json, "customers[].name")`

Previously, JMESPath validation only ran inside Code nodes. If you use n8n-mcp's `validate_workflow`, upgrade to v2.87.0 or later; it statically checks `$jmespath()` calls inside expressions and reports these errors before runtime.

Question 4

Why does the n8n Python Code node fail with NameError for _input or _json and block json/datetime imports?

Accepted Answer

n8n 2.x uses the native Python runtime (`language: "pythonNative"`), not the removed Pyodide "Python (Beta)" runtime. Update Code node scripts as follows:

1. Replace `_input.all()` with `_items` in All Items mode:
   ```python
   return [{"json": item} for item in _items]
   ```
2. In Each Item mode, use `_item` instead of `_json`:
   ```python
   return {"json": _item}
   ```
3. Access data with dictionary syntax only. `_input`, `_json`, `_node`, `_now`, `_today`, and `_jmespath` are unavailable.
4. Do not use `import json`, `import datetime`, or other stdlib imports unless your self-hosted task runner explicitly allowlists them. n8n Cloud allows no imports by default.
5. Avoid `class`, `type()`, `getattr()`, `hasattr()`, dunder access, and `global`; use `nonlocal` where needed.
6. Return values must match the mode: All Items accepts a list of `{"json": ...}` objects, a list of plain dicts, or one dict; Each Item accepts one dict. Returning `None` drops an item, while returning a list errors.

If self-hosted Docker reports `Python runner unavailable: Python 3 is missing from this system`, deploy the `n8nio/runners` sidecar; otherwise Python Code nodes cannot run.

Question 5

How to fix n8n 403 "insufficient_api_key_scope" or "insufficient_permissions" when updating a published workflow?

Accepted Answer

In n8n 2.39+, `PUT /workflows/{id}` saves changes as a draft but returns 403 when the caller cannot publish. The published workflow stays live. Grant the API key the `workflow:activate` scope and ensure the user has `workflow:publish` permission on that workflow, then retry the update. If using n8n-mcp, upgrade to v2.87.0 or later; these responses are reported as `PUBLISH_FORBIDDEN` instead of the previous misleading partial-update message. With `n8n_update_full_workflow`, the draft is retained; with `n8n_update_partial_workflow`, the draft is rolled back. Retrying with the same restricted credentials will create another draft and will not publish.

n8n-mcp

Overview

README Preview

FAQ (5)

同类型项目

vscode

FreeDomain

stable-diffusion-webui

paperclip