What the data for this category looks like
The core data for delivery effect optimization comes from advertising platform data, attribution systems, CRM leads, and landing page data, with a daily or real-time update frequency. Daily produced documents include daily reports, weekly reports, price adjustment records, material test sheets, and optimization reviews. Data fields include impressions, clicks, conversion cost, budget balance, etc. Units involve yuan, times, cost per thousand impressions, etc. The data structure has a relatively high degree of standardization, with fixed core fields, making batch parsing and calling convenient.
What constraints these characteristics impose on the tool calling and plugins link
Since data sources are diverse and updates are real-time, tool calling needs to adapt to the automatic selection logic for multiple data sources to avoid result deviations caused by incorrect data source matching. The standardized structure of daily documents requires plugins to support parsing of fixed fields, and cannot be compatible with non-standardized custom field formats. Real-time updated data requires that the timeout setting for tool calling matches the data delay window, while also considering the API rate limiting rules of advertising platforms to avoid concurrent calls exceeding limits. In addition, the need to integrate multiple types of data requires tool calling to simultaneously associate outputs from multiple data sources such as advertising platforms and attribution systems.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
toolChoice | auto | Adapts to the automatic tool selection logic for multiple data sources, covering data sources such as advertising platforms and attribution systems |
functionCall | {"name": "query_ad_performance"} | Binds the dedicated data query tool for delivery effect optimization, clarifies the calling target |
| Tool call timeout | 600 seconds | Matches the real-time data update delay window, avoids empty results caused by unsynchronized data |
| Concurrent call limit | First 10 calls per second | Adapts to the API rate limiting rules of advertising platforms, prevents triggering rate limits and returning empty results |
| Document parsing character limit | 800–1200 characters | Adapts to the single-page content length of daily reports and price adjustment records, avoids parsing truncation |
| Multimodal support | Enabled | Adapts to the image material parsing requirements in material test sheets |
Three Common Misconfigurations
- When concurrent call volume reaches 2-3 calls per second or higher, tool calling returns none. The cause is that a reasonable concurrent call limit is not set, exceeding the API call limits of advertising platforms.
- Unable to connect when calling an interface. The cause is that the
toolChoiceparameter is not configured correctly, leading to abnormal tool selection logic that cannot match the target data source. - MCP function parameter passing often fails. The cause is that the fixed field format of daily documents is not aligned, leading to parameters not being correctly recognized.
How to Confirm Proper Configuration
- Initiate a single tool call to verify that the returned results include core delivery indicator fields such as impressions and conversion cost.
- Initiate concurrent calls at 10 calls per second to confirm that no empty results or error messages are returned.
- Upload a price adjustment record document to verify that the tool can correctly extract fixed fields such as price adjustment amount and adjustment date.
- Pass a single test image to confirm that the tool can correctly parse material-related information in the image.
The provided values are common starting points and should be evaluated against relevant samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.