What the data for this category looks like
Joint-stock bank financial report data mainly comes from official disclosure channels, including the official platform of the China Banking and Insurance Regulatory Commission, official websites of bank annual and quarterly reports, and stock exchange announcement systems. Annual reports are disclosed collectively by the end of April each year. Quarterly reports are released within one month after the end of each quarter. The document structure includes two parts: standardized structured statements and notes. Structured statements cover the balance sheet, income statement, and cash flow statement. Fields include bank-specific indicators such as core tier 1 capital adequacy ratio, non-performing loan ratio, and provision coverage ratio. Units are mostly 100 million yuan and percentage.
What constraints do these characteristics impose on the tool calling and plugins link
The unique characteristics of joint-stock bank financial reports impose multiple constraints on the tool calling and plugins link. Multiple sources of disclosure data require plugins to connect to multiple public channels including the China Banking and Insurance Regulatory Commission, exchanges, and bank official websites. They also require verification of the compliance of data disclosure time. A single annual financial report can reach tens of thousands of characters. Tool calling must adapt to long text chunk parsing and result merging logic. The indicator fields specific to banks, such as capital adequacy ratio and non-performing loan ratio, differ from general financial report fields. Dedicated field mapping rules must be configured. The concentrated disclosure cycle creates demand for batch financial report processing. Tool flows must support scheduled batch scheduling.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Single joint-stock bank annual financial report has a large length; 600 seconds covers the complete parsing process |
maxToolCall | 8–12 times | Financial report analysis requires multiple steps of tools including multi-data source calls, field verification, and result integration; 8-12 times covers the complete business logic |
FIELD_MAPPING_RULES | Configure bank-specific indicator mapping table | Joint-stock bank financial reports include dedicated fields such as core tier 1 capital adequacy ratio; custom mapping to general fields required for analysis is needed |
MCP_SERVER_DEPLOY_PATH | /opt/fastgpt/mcp | Standard deployment path for version 4.9.6, adapts to the tool flow scheduling requirements of batch financial report processing |
tokenCalculationMode | Calculate based on actual token count of a single tool call | Matches the actual resource consumption of user calls, complies with public token billing logic |
RECALL_FILE_MAX_SIZE | 1000 MB | Adapts to the maximum document size of a single bank financial report, avoids parsing failure |
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, and it is recommended to test on your own samples before finalizing.
Three Common Mistakes
- Phenomenon: Tool flow cannot be triggered after deploying the MCP server, returning a 503 status code. Cause: The
MCP_SERVER_DEPLOY_PATHis not configured to the standard deployment path of version 4.9.6, causing the service to fail to complete registration and scheduling. - Phenomenon: The AI fails to recall specified financial report knowledge base content during tool calling, or the recalled content conflicts with tool logic. Cause: The
toolDataSourceWhitelistis not configured to limit legal data sources, causing confusion in the priority of knowledge base recall and tool calling data sources. - Phenomenon: Tool calling fails when calling the gemini model, returning the error message "model does not support tool calling". Cause: The tool calling support configuration for the corresponding model is not enabled, or the model version that supports tool calling is not selected.
How to Confirm the Configuration is Correct
- Enter the tool flow editing page, check whether the
maxToolCallconfiguration value matches the current setting, and confirm that the tool call frequency limit meets expectations. - Upload a test joint-stock bank financial report, check whether the parsed fields include bank-specific indicators, and confirm that
FIELD_MAPPING_RULEStakes effect. - Start the MCP server service, check whether the console log shows that the service registration is successful, and confirm that the deployment path configuration is correct.
- Trigger a tool call, check the token consumption log, and confirm that the
tokenCalculationModecalculation logic matches the preset rules.
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.