What the data for this category looks like
The data used for meeting minutes organization primarily comes from on-site meeting recording transcriptions, meeting notes submitted by attendees, and original attendance records exported from meeting management systems. Update frequency follows the meeting cycle. Minutes data for a single meeting is initially generated after the meeting concludes. In real-time meeting scenarios, data is updated on a per-minute basis. The document structure includes six fixed core fields: meeting topic, meeting time, attendee list, discussion topic summary, agreed matters, and follow-up action items. The action item field must be linked to responsible persons and deadlines. The length of each field varies significantly: some discussion topic summaries are lengthy, while individual action item descriptions are concise and clear.
Constraints on Tool Calling and Plugins Derived from These Characteristics
Diverse data sources and inconsistent formats for meeting minutes require tool calling plugins to support automatic adaptive parsing of multi-source inputs, to avoid extraction errors caused by accent interference in transcriptions or differences in handwritten note formats. Differences in update frequency require plugins to have configurable trigger timing, supporting batch processing after a meeting concludes or incremental updates during real-time meetings. The fixed core field structure requires tool calling to specifically extract designated content such as agreed matters and follow-up action items, to avoid redundant processing of invalid information. Differences in content length require the plugin's context processing capability to be flexibly adjusted to adapt to minutes texts of varying lengths, preventing processing failures due to single text segments exceeding tool calling limits.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
toolChoice | auto | Adapt to tool calling requirements across different scenarios, no manual specification of individual tools needed |
functionCall | Call by field category | Target different tools to extract corresponding content for the fixed field structure of meeting minutes |
| Trigger timing | Trigger within 1 hour after the meeting concludes | Match the generation cycle of most meeting minutes, avoid resource consumption from real-time processing |
| Multi-tool call order | Extract attendees first, then agreed matters, then action items | Follow the logical generation order of meeting minutes, ensuring prior content provides context for subsequent tool calls |
| Timeout setting | 300–600 seconds | Adapt to the time required for multi-source data parsing, avoid call failures due to long text processing |
| Maximum character count per call | 8000–12000 characters | Adapt to the typical length of meeting minutes texts, avoid exceeding tool processing limits |
Three Common Configuration Mistakes
- Symptom: FastGPT fails to load MCP tool plugins after startup, with logs showing connection timeout. Cause: After upgrading fastgpt-mcp-server, the service address and port parameters in the plugin configuration were not updated synchronously, preventing normal service connectivity.
- Symptom: The LLM calls tools in a disordered order during a single query, and the extracted meeting minutes fields are logically reversed. Cause: No fixed multi-tool call order was configured, causing the LLM to randomly select tool execution order.
- Symptom: The action item field is missing or empty in the results returned by tool calls. Cause: The targeted field extraction configuration for
functionCallwas not enabled, so the tool did not call for the core fields of meeting minutes.
How to Verify the Configuration is Correct
- Check if the
toolChoiceparameter on the plugin configuration page is set toautoor a specified tool combination, and confirm the parameter has been saved and takes effect. - Upload a test meeting minutes text, trigger a tool call, and check if the returned results include core fields such as meeting topic and agreed matters.
- Review the tool call logs to confirm that multi-tool calls execute in the preset order, with no timeout errors.
- Compare the original meeting data with the tool extraction results to confirm that the extracted fields conform to the standard structure of meeting minutes.
The values provided are common starting points and should be evaluated against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.