Hermes Agent shows the tool is enabled, but when answering, it never uses the tool; it checks the Tools Runtime first rather than directly suspecting the model is "not smart." For a tool to be truly called, it must meet at least three requirements: successful runtime registration, permission for the current session, and clear task prompts requiring tool intervention. If any link is missing, the model may only answer in text.
First, let's see if it's a registration issue
If the tool comes from an MCP, plugin, or custom provider, first confirm that the service process has started. Hermes can see the tool schema and it has not been skipped due to environment variables, paths, or permission failures. Many cases where the tool is enabled but not used actually don't enter the current session, only written in the configuration file.
Then check whether the task really requires tools
Users only ask "What might be the reason for this error?" The model can analyze purely; If you want it to read files, run commands, or search web pages, you need to clearly state the actionable actions, such as "Please check the current project's package.json and test output." Hermes Agent doesn't mean you have to call a tool just because you see it. A good tool should serve the task, not just make it seem noticeable.
Finally, look at the model and strategy
Some models have weak support for tool call formats, and some session policies require high-risk tools to be approved first. What appears to be "not called" may actually be due to incompatible tool schemas, approval blocking, or system prompts writing the scope of action too conservatively.
It's best to do a minimal test during troubleshooting: keep only one simple tool and let it accomplish a task that requires calling that tool to complete. If this test works, add MCP, file tools, and browser tools one by one. Installing a dozen tools at once and then troubleshooting basically only makes the problem more obscure.
