Tool Use — Extend Hermes with Any Tool
Key Points
- ✓MCP (Model Context Protocol) support
- ✓Custom tool definitions in YAML
- ✓HTTP API tools
- ✓CLI command tools
- ✓Tool chaining for complex workflows
- ✓Community tool library
How It Works
- 1Define tools in config or install from marketplace
- 2Hermes auto-discovers available tools
- 3Tools are called when relevant to your request
- 4Chain multiple tools for complex workflows
Real-World Use Cases
MCP Server Integration
Connect any MCP-compatible server — Cursor extensions, VS Code plugins, Claude Desktop tools — directly to Hermes. Browse available MCP tools, install them, and Hermes auto-discovers and uses them when relevant to incoming tasks.
Custom HTTP API Tools
Define your internal API as a YAML tool spec: endpoint, parameters, auth headers, response schema. Hermes calls it like a native tool, handles authentication, parses the response, and uses the results in multi-step workflows — without writing integration code.
CLI Command Tools
Wrap any CLI tool as a Hermes tool: kubectl, terraform, aws cli, custom scripts. Define input parameters and expected output format; Hermes handles argument construction, execution, and output parsing automatically.
Multi-Tool Pipeline Orchestration
Chain tools for complex workflows: browser automation extracts data, code execution transforms it, an HTTP API tool stores it, and a notification tool reports completion. Hermes builds the pipeline automatically from the task description.
Under the Hood
Hermes tool selection is fully automatic — at each step in a task, the agent reasons about which tool to reach for based on the current goal, available tools, and prior outcomes. The tool selection process uses a two-level discovery approach: first, the agent scans tool descriptions (~3k tokens for the full list) to narrow candidates; then it inspects the full specification of candidate tools before calling them. This progressive disclosure keeps token overhead proportional to the number of tools actually used, not the total tool count.
MCP (Model Context Protocol) support allows Hermes to act as both an MCP client (consuming tools from MCP servers) and an MCP server (exposing Hermes conversations as a tool source for Claude Desktop, Cursor, or VS Code via hermes mcp serve). The bidirectional MCP support means Hermes integrates cleanly into existing AI-enhanced development environments rather than requiring you to abandon your current toolchain.
Custom tool definitions use a YAML schema that covers authentication patterns (bearer tokens, API keys, OAuth), parameter types (string, number, boolean, array, object with schema), response handling (JSON path extraction, regex capture, full response passthrough), and error handling (retry logic, fallback behavior). Tools can be defined inline in config.yaml or as separate files in a tools/ directory. The community tool library at the Skills Hub includes pre-built definitions for popular APIs, so common integrations are a single install away rather than a YAML authoring exercise.