This document is based on FastGPT v4.17.0.
What the data for this category looks like
The core data for on-site outsourced operations and maintenance services comes from the customer’s ITSM system, on-site monitoring platform, asset inventory ledger, and on-duty records. Daily produced documents include on-site daily reports, service desk tickets, inspection reports, and SLA monthly reports. Data updates are primarily daily, with summary statistics generated monthly. The fields of various documents include ticket numbers, inspection item numbers, service response durations, on-site staff attendance records, etc. There is no unified standardized field format, and adaptation to different customers’ system structures is required.
How these characteristics impose constraints on tool calling and plugins
The data sources for on-site outsourced operations and maintenance services are scattered and have inconsistent formats. This requires that tool calling plugins support multi-source data parsing and field mapping, and adapt to custom fields of different customers. Data update frequency is mainly daily or monthly, so tool calling frequency should not be too high, to avoid interface rate limiting caused by frequent pulling of customer system data. Tool calling results in on-site scenarios must comply with customer SLA requirements, sensitive data must be hidden, and the termination command for tool calling must adapt to the on-site operation habits of on-site engineers without affecting the customer’s existing operations and maintenance processes.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
toolChoice | none | Hides the original execution results of tool calls, only returns organized data related to on-site operations, which complies with customer confidentiality requirements |
functionCall | Auto-trigger | Adapts to the daily operations of on-site engineers, no manual tool call triggering is required, automatically pulls the latest tickets and inspection data |
| Single tool call timeout | 600 seconds | Adapts to the interface response duration of the customer’s ITSM system, avoids tool call failures caused by timeouts |
| MCP server tool termination command trigger | Triggered by on-site staff input commands | Complies with the on-site operation habits of on-site engineers, can terminate abnormally executing tools at any time |
| Tool call context limit | Top 5 entries | Adapts to the typical data volume of on-site daily reports and inspection reports, avoids tool call interruptions caused by context overflow |
| Tool call result output switch | Off | Complies with information confidentiality requirements in on-site scenarios, only returns organized analysis results to on-site engineers |
Three Common Mistakes
- Phenomenon: The original execution result is still displayed in the conversation interface after a tool call completes. Cause:
toolChoiceis not set tonone, or the tool call result output switch is not turned off. - Phenomenon: Unable to send a termination command to the MCP server-side tool. Cause: The termination trigger interface for the MCP tool is not bound in the plugin configuration, or correct trigger rules are not configured.
- Phenomenon: The model randomly interrupts output when calling a tool. Cause: A reasonable single tool call timeout period is not set, or the context length exceeds the configured limit, leading to automatic truncation.
How to Verify the Configuration is Correct
- Initiate a tool call, check that the conversation interface only displays organized on-site data, with no original tool execution logs.
- Input a termination command during tool execution, check that the MCP server-side tool stops executing and returns a termination prompt.
- Simulate a single tool call timeout scenario, check that the system triggers a timeout error and terminates tool execution.
- Adjust the tool call context limit to
Top 3 entries, initiate a call, and check that the number of returned data entries matches the configured value.
The values provided here 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 2026-10-19.