Agent integrations
The repository ships an agent skill and an MCP bridge plugin so coding agents can call a wickworks server you run. Neither contains the analysis engine. Start the server first (Install and run), then point the agent at its URL.
| Piece | Where | What it does |
|---|---|---|
| Skill | .agents/skills/wickworks/ | Teaches an agent the REST and MCP API, the indicator names and the errors |
OpenClaw plugin @psyb0t/wickworks | .agents/plugins/wickworks/ | A stdio-to-HTTP MCP bridge for clients that only run local stdio servers (MIT licensed) |
| Claude Code and Codex manifests | .agents/.claude-plugin/, .agents/.codex-plugin/ | Package the skill as a plugin for those clients |
The skill works in any agent that reads .agents/skills/, such as peen.
Claude Code
claude plugin marketplace add psyb0t/agents
claude plugin install wickworks@psyb0t
Claude Code asks for the wickworks URL and, if a reverse proxy in front of wickworks requires one, a bearer token. It stores the token in your OS keychain.
Claude Code also speaks remote MCP, so you can connect the tools without the plugin:
claude mcp add --transport http wickworks http://localhost:8000/mcp
Codex
codex plugin marketplace add psyb0t/agents
codex plugin add wickworks@psyb0t
Installed through the marketplace, the skill is invoked as $wickworks:wickworks. Codex also picks the skill up without any install in a repository that contains .agents/skills/; there it is invoked as $wickworks.
OpenClaw and other stdio clients
The skill is published to ClawHub with every release:
openclaw skills install @psyb0t/wickworks
The plugin bridges a stdio MCP client to the server’s /mcp/ endpoint through mcp-remote:
openclaw plugins install clawhub:@psyb0t/wickworks
Then set these in the plugin’s environment:
| Variable | Required | Description |
|---|---|---|
WICKWORKS_URL | yes | Base URL of your server, such as http://localhost:8000. The bridge appends /mcp/ |
WICKWORKS_TOKEN | no | Sent as Authorization: Bearer <token>. Only needed when a reverse proxy in front of wickworks checks it; wickworks itself has no auth |
The bridge exits with an error message on stderr when WICKWORKS_URL is unset. It writes nothing else of its own, because stdout carries the MCP protocol, and it never logs the token.
The bridge runs on Node.js 18 or newer. Any MCP client that starts stdio servers can run its wickworks-mcp command the same way.
Tools the agent gets
Through MCP, by plugin or directly: health, list_indicators, metadata and compute. See MCP server for arguments and results. Through the skill alone, the agent calls the REST API with curl; the skill lists docker (to run the server) and curl as the programs it needs.