Failed to transform agent message: PluginInvokeError:

[Binnn-MX]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:)
• Please do not modify this template :) and fill in all the required fields.

1. Is this request related to a challenge you're experiencing? Tell me about your story.

Dify version
1.3.0

Cloud or Self Hosted
Self Hosted (Source)

Steps to reproduce
dify version: 1.3.0 source

agent node error

Using the custom MCP, the operation fails most of the time, but sometimes it can succeed.

Run failed: Failed to transform agent message: PluginInvokeError: {"args":{},"error_type":"JSONDecodeError","message":"Unterminated string starting at: line 1 column 80 (char 79)"}

2. Additional context or comments

No response

[dosubot[bot]]

Your error is caused by malformed JSON—specifically, an unterminated string—returned from your custom MCP or plugin. This usually happens when the MCP outputs incomplete or invalid JSON, which Dify tries to parse during agent message transformation or tool invocation. Intermittent success suggests your MCP sometimes returns valid JSON, but often doesn't.

To fix this, ensure your MCP always returns well-formed JSON. Double-check that any string fields are properly quoted and closed, and that the overall structure matches what Dify expects. For agent plugins, each context item should include content, title, and metadata fields, not just a result key. If your MCP outputs only a result, wrap it in the required structure before returning it. See a relevant example and explanation here.

Also, review your MCP's logs and outputs for cases where the JSON might be truncated or malformed. If you’re streaming responses, make sure the stream is always completed and closed properly. If you recently changed your MCP or plugin code, check for any code paths that could return incomplete data.

If you’re using a plugin or model that’s known to have compatibility issues (like Tongyi in function calling mode), try switching to a different provider (such as OpenAI API) or ensure all required configuration fields are present, especially if you see errors about missing fields or descriptions example.

Finally, make sure your Dify core and plugin-daemon versions are aligned, as version mismatches can cause similar issues. If you need to debug further, add logging to capture the exact JSON output from your MCP right before it’s returned to Dify, and validate it with a JSON linter.

Let me know if you want more specific debugging steps or code examples!

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[Modas-Li]

same issue:
api-1 | 2026-02-25 01:39:36.902 ERROR [Thread-22942 (_generate_worker)] [workflow_entry.py:111] - Unknown Error when workflow entry running
api-1 | Traceback (most recent call last):
api-1 | File "/app/api/core/workflow/workflow_entry.py", line 107, in run
api-1 | yield from generator
api-1 | File "/app/api/core/workflow/graph_engine/graph_engine.py", line 270, in run
api-1 | raise self._graph_execution.error
api-1 | RuntimeError: Failed to transform agent message: req_id: 18944f37d2 PluginInvokeError: {"args":{},"error_type":"JSONDecodeError","message":"Expecting property name enclosed in double quotes: line 1 column 2 (char 1)"}