What the data for this category looks like
The data for survey and design declaration materials and file review is primarily sourced from construction drawing documents submitted for project review, qualification declaration archives, and project completion files. Updates trigger with individual project review submission workflows, with no fixed cycle. Most documents are multi-page PDFs and CAD format attachments, paired with structured form data. Fields include project number, survey scope, review opinions, and qualification level. Building area and review opinion clause numbers are common units and identification items.
What constraints these characteristics impose on tool calling and plugins
Mixed-format document structures require tool calling plugins to support PDF text extraction and CAD drawing metadata parsing, avoiding parsing failures caused by only adapting to a single document format. Unstructured review opinions and structured form fields coexist, so tool calling must trigger both text semantic extraction and structured field mapping, preventing missing key parameters. Single-project data updates with the review submission process require plugins to bind the corresponding project’s data source. Global caching should be avoided to ensure the latest version of files and declaration materials is read each time the tool is called. Large CAD attachment sizes impose additional constraints on the plugin’s file upload threshold and parsing timeout settings.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
toolChoice | auto | Adapts to scenarios where structured and unstructured content are mixed in declaration materials, automatically selecting the appropriate calling mode |
functionCall | Enabled | Supports accurate mapping of structured fields from survey and design files, such as project number and review opinion clause numbers |
| Tool calling timeout | 120–180 seconds | CAD attachment parsing and multi-document merging processing require longer time, calibrated based on actual testing |
| Maximum number of documents per call | First 3 | Single-project declaration materials usually do not exceed 3 core documents, to avoid context overflow |
| Multi-format file support toggle | Enabled | Covers the two common data sources: PDF construction drawing files and CAD attachments |
| Structured field matching threshold | 0.85–0.9 | Survey and design field terms are highly professional, requiring a high matching degree to avoid misjudgment |
Three common mistakes
- Phenomenon: When calling a third-party patent query MCP tool, the tool returns that the enterprise has no patent information records, but the tool set test can return results normally. Cause: The enterprise subject parameter corresponding to the project was not correctly bound during tool calling, leading to deviation in the retrieval scope.
- Phenomenon: When using the official Markdown to file conversion plugin, an upload file failure message is displayed. Cause: The multi-format file support toggle was not enabled, or the uploaded file size exceeded the plugin's preset 100 MB threshold.
Reference only; verify against the target FastGPT release and deployment documentation.
How to confirm the configuration is correct
- Upload a single CAD attachment and PDF declaration material, verify that the plugin can normally parse and extract structured fields.
- Call the tool set test and formal calling components, compare the consistency of returned results to confirm that parameters are correctly bound.
- View plugin logs to confirm there are no MongoDB syntax errors or timeout errors.
- Adjust the tool calling timeout time to verify whether parsing of large-volume CAD files works properly.
The values provided are common starting points and should be measured 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.