Lesson 5 of 6
Teach it new tricks: skills and MCP servers
Browse the community skills hub, install a skill, connect a filesystem MCP server, and watch Hermes reach for both automatically.
Hermes ships with 91 built-in skills and access to 500+ more in the community hub. MCP servers add live connections to external tools — GitHub, filesystems, APIs — using an open standard that works across any MCP-compatible agent.
This lesson installs one of each so you can see how Hermes decides when to reach for them.
Skills: procedural memory
A skill is a markdown file with YAML frontmatter that tells Hermes when to invoke it (the trigger conditions) and how to do the thing (the procedure). Hermes reads only the frontmatter at session start — the body loads on-demand, so skills don't bloat your context window.
Browse the hub
hermes skills browse
This opens a paginated TUI browser. Use arrow keys to navigate categories: Software Development, Creative, MLOps, Research, and more. The community hub at agentskills.io has the same catalog in a browser.
Search and install
hermes skills search "github"
hermes skills install github-pr-review
Replace github-pr-review with whatever you found. Install anything that looks useful for your current work.
Always audit after installing community skills
hermes skills audit
hermes skills audit is a manual command — it doesn't run automatically on install. Run it after installing any community skill. Hermes flags anything that could execute shell commands in unexpected ways or access files outside your project. This is a one-minute step that's worth the habit.
Run hermes skills audit after every community skill install. The skills registry is open — anyone can publish. Auditing takes seconds and Hermes will flag anything suspicious.
Test it
Open a chat and give Hermes a task that should trigger the skill you installed:
Search the skills hub and recommend 3 skills that would help someone building a Next.js app. Show me what each one does before installing anything.
Watch the output — if Hermes reaches for a skill automatically rather than just writing freeform text, the skill system is working.
Hermes learns over time which skills to reach for. After a few sessions it will start invoking installed skills automatically when the task matches — you don't have to name them explicitly.
MCP: connect external tools
MCP (Model Context Protocol) is an open standard for exposing tools to any compatible agent. Hermes supports any MCP server — once added, its tools appear alongside Hermes's built-in tools.
Best first MCP server: filesystem
The filesystem MCP server lets Hermes read and write files in a specific directory — useful for working on a project without giving it access to your entire disk.
Add it by editing ~/.hermes/config.yaml:
hermes config edit
Add this under a mcp_servers: key (create the key if it doesn't exist):
mcp_servers:
project_fs:
command: "npx"
args: ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/your-project"]
Replace the path with your actual project directory. Save the file, then reload MCP in a running chat with:
/reload-mcp
Or restart Hermes. Confirm the server is connected:
hermes mcp list
Test the MCP connection
hermes mcp test project_fs
Then ask Hermes to use it:
Using the project filesystem tool, list the files in the root of my project and tell me what framework it looks like and where the main entry point is.
GitHub MCP (second recommendation)
If you work with GitHub, the GitHub MCP server is the next most useful:
mcp_servers:
github:
command: "npx"
args: ["-y", "@modelcontextprotocol/server-github"]
tools:
include: [list_issues, create_issue, search_code, list_pull_requests]
Use tools.include to allowlist only the operations you actually want — this prevents Hermes from accidentally closing issues or merging PRs.
Skills vs MCP — what's the difference? Skills are agent-written procedures (like a recipe: do this, then this, then this). MCP servers are live API connections (like a phone line to GitHub). They're complementary — a skill might use an MCP tool as one of its steps.
The curator: background skill improvement
The curator is a background process that reviews your installed skills, improves ones that have gotten stale, and turns repeated patterns from your sessions into new skills.
hermes curator status # is it running?
hermes curator run # trigger a manual review
hermes curator pin github-pr-review # freeze a skill you don't want auto-updated
The curator runs automatically in the background. If it updates a skill in a way you don't like, use hermes curator pin <skill> to freeze it, or hermes checkpoints to roll back.
See your activity
After a few sessions with skills active:
hermes insights
This shows token usage, cost, and activity broken down by platform and skill. Useful for catching runaway cron jobs and understanding where your token spend is going.
The final lesson explains how memory works across all of this — and how to build a long-term knowledge base that grows with you.