When you start configuring a Hermes Agent profile, the most common friction points are picking the right provider, turning off unnecessary toolsets, and wiring up MCP servers without breaking the config file. First, decide where your model will come from. If you need broad model access, choose OpenRouter; for the latest Nous research models, pick Nous Portal; if you already have an NVIDIA GPU setup, select NVIDIA; otherwise, the OpenAI‑compatible custom option lets you point to any local or third‑party endpoint.
Next, handle built‑in toolsets. The UI shows that setting the toggle to Off writes an empty list to agent.disabled_toolsets. Use this to strip out tools you never need—like code execution or web search—so the agent starts lighter, faster, and less prone to unwanted side effects.
For MCP servers, the selector lets you add Filesystem, GitHub, or HTTP integrations. After choosing a type, fill the placeholder with the actual path or URL (e.g., /home/user/projects for Filesystem or https://api.github.com for GitHub). Clicking + Add stores each entry as a key‑value pair under mcp_servers. Remember that the configuration expects a map, so each server name must be unique; duplicate names will overwrite previous entries.
Once you’ve filled Provider, disabled toolsets, and MCP servers, the preview pane shows the target file path—typically ~/.hermes/profiles/researcher/config.yaml. Press Copy to place the rendered YAML on your clipboard, then paste it into your editor or directly into the profiles directory. Validate the file with a YAML linter before launching the agent to catch indentation errors early.
By following these three straightforward steps—provider selection, toolset pruning, and MCP server mapping—you avoid configuration drift, reduce startup latency, and keep your Hermes Agent focused on the tasks that matter.