What the data for this category looks like
Traditional Chinese medicine (TCM) research report data primarily comes from professional pharmaceutical industry databases, public documents from the National Pharmacopoeia Committee, and annual reports from TCM industry associations. The regular update cycle is quarterly, with ad-hoc updates triggered by revisions to the national pharmacopoeia, new drug approvals, or major industry policy releases. Each individual document includes sections such as original source identification, decoction pieces specifications, clinical compatibility, pharmacologically active ingredient data, and national pharmacopoeia standard fields. Content determination data is often measured in mg/kg or g/100g, and clinical cases note specific usage dosages and treatment courses.
How these characteristics constrain tool calling and plugin configuration
These data characteristics directly constrain the configuration logic for tool calling and plugins. Professional terminology and standardized fields require tool calling to precisely match specified dimensions, to avoid irrelevant results from generalized retrieval. The multi-dimensional data structure requires plugins to support filtering recall results by field, such as categorizing by pharmacopoeia inclusion status or active ingredient type. The variable update cycle requires plugins to support incremental data source pulling, while reserving an interface for manually triggering full updates. Long-form content requires tool calling nodes to be configured with a reasonable context window, to avoid truncating core information.
Configuration Recommendations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
toolCallMaxTurns | 3–5 | TCM research report retrieval often requires multiple rounds to verify field accuracy; excessive turns increase overall latency |
toolCallTimeout | 120 seconds | TCM research report documents are lengthy, and integration with multiple data sources such as pharmacopoeia databases and market databases requires adapting to the time consumption of multiple interface calls |
Recall Field Filter | ["药典编号", "活性成分含量", "临床适应症"] | Core retrieval dimensions for TCM research reports are standardized fields; filtering non-essential fields improves recall precision |
maxContext | 8000–12000 characters | Single TCM research report documents have lengthy valid content, requiring sufficient context for tool calling to generate accurate instructions |
Incremental Sync Cycle | 7 days | The regular update cycle for TCM research reports is quarterly; weekly incremental synchronization covers the timeliness requirements of ad-hoc updates |
Similarity threshold | 0.75–0.85 | TCM terminology has aliases; a threshold that is too low introduces irrelevant results, while a threshold that is too high misses precise matches for synonymous terms |
The parameter values provided on this page are common starting points for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require specific analysis; it is recommended to test on your own samples before finalizing.
Three Common Misconfigurations
- Symptom: The model output interrupts mid-execution after a tool calling node runs, with no subsequent results. Cause: The
toolCallTimeoutparameter is not configured, or its value is shorter than the total time required for multi-data source interface calls, resulting in forced task termination. - Symptom: A
403 Forbiddenerror is triggered when calling an external document processing tool in a workflow. Cause: The access whitelist for the corresponding tool is not added in the plugin configuration, or correct API access permissions are not configured. - Symptom: When debugging a tool calling node, the interface displays two separate thought processes. Cause: Both the
toolCallAutoReviewswitch and the automatic supplement logic for the default system prompt are enabled, resulting in repeated triggering of the thought generation process.
How to Confirm Proper Configuration
- Trigger a full-field retrieval for a single TCM research report, and verify that the results returned by the tool call include the preset specified fields.
- Check the workflow execution log, and verify that the effective duration of the
toolCallTimeoutparameter matches the configured value, with no timeout error records. - Call the workflow via API, pull the tool call logs during execution, and verify that the returned MCP tool call parameters match the configured items.
- Debug the tool calling node, confirm that only one thought process is displayed, with no logically repeated trigger records.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.