AI Toolbox
Every AI tool your CLIs use — agents, skills, commands, rules, and MCP servers — lives in the Hermes toolbox. Author, install, and configure them from your desktop or your phone; wherever you work from, the tools themselves execute on your paired desktop.
Marketplace
Install AI tools without authoring them yourself. The marketplace has two sections — community repositories and official plugins — so you can pull in third-party tools or manage Claude-native plugins from the same place.
- Community repos: search and install AI tools from GitHub repositories into your project, at user scope or project scope.
- Official plugins: install, uninstall, and toggle Claude-native plugins.
- Tools install for whichever AI CLIs you've selected.
- Browse, search, and preview any tool before you install it.
What you need
An AI CLI installed, and an internet connection for marketplace access.
Hermes Toolkit
Skip authoring your conveyor tooling from scratch. The Hermes toolkit is a built-in set of hermes_-prefixed entities tuned for the conveyor pipeline, and an onboarding wizard installs all of them into your project in one pass.
- Built-in
hermes_-prefixed toolkit entities optimized for the conveyor pipeline. - Specialized agents including
hermes_planner,hermes_impl-auditor,hermes_plan-validator,hermes_qa-tester,hermes_task-analyzer,hermes_deep-planner,hermes_deep-debugger, and more. - Skills including
hermes_commit,hermes_merge-worktrees, and an audit skill also namedhermes_impl-auditor. - A two-step onboarding wizard installs every
hermes_-prefixed entity into your project. - New kanban boards default their conveyor agents to
hermes_agents once the toolkit is installed. - Toolkit entities are version-managed and updated through the marketplace.
What you need
The Hermes desktop app, with toolkit onboarding completed.
MCP Servers
Everything above is a surface you use to manage tools yourself. MCP servers are different: they're what your AI agents connect to directly while they work. Hermes ships four built-in MCP (Model Context Protocol) servers that give agents structured, scoped access to your project.
hermes_db
Database access for AI agents. Query connected MySQL, Elasticsearch, and SQLite databases. Read-only by default, with a per-connection write guard for explicit opt-in, and every query runs inside a read-only transaction for safety.
hermes_github
GitHub operations for AI agents. List, read, create, and comment on pull requests, and request or submit code reviews. Merge is never exposed through this server — merging always stays human-gated.
hermes_requests
API request execution for AI agents. List collections and saved requests, then execute stored requests with resolved variables. Per-collection access mode controls exactly what agents are allowed to do.
hermes_tasks
Task queue and conveyor state for AI agents. Read-only access to task queue entries and kanban conveyor state, including stage history and output transcripts, so agents understand where a project stands.
These servers run as stdio child processes on your desktop. AI CLIs discover and connect to them automatically once they're configured in your project's toolbox.
What you need
An AI CLI with MCP support (Claude Code), and the relevant service connections configured — database connections, GitHub, and so on.
Capability Matrix
Not every AI CLI supports every toolbox feature. A per-platform probe capabilities button shows exactly what each one supports, so you know what to expect before you invest time authoring or installing tools it can't run.
- Per-platform probe capabilities button.
- Shows what each AI CLI supports: tool authoring, MCP servers, skill execution, agent dispatch.
- Helps you understand which toolbox features are available under your current CLI configuration.
What you need
At least one AI CLI installed.