Give AI agents the platform knowledge and live tools they need to build Apps and integrations on epilot and to configure the platform end to end.
The plugin appears as epilot and keeps epilot-core as its stable
technical identifier. It combines portable platform
knowledge with live tools:
- epilot MCP for current documentation, OpenAPI discovery, entity schemas, and connected-organization configuration;
- Volt UI MCP for current components, props, and design tokens;
- a thin platform guide plus focused skills for Apps (including native interface design with Volt UI), integrations, end-to-end platform configuration, journey building, and workflow building;
- a deterministic local inspector for epilot App manifests and build inputs.
The toolkit intentionally excludes Lima, SAP, and other ERP-specific behavior. Vendor implementations belong in separately installable plugins.
Add this repository as a marketplace:
codex plugin marketplace add epilot-dev/agent-toolkit-for-epilotThen open /plugins in Codex and install epilot-core.
Run these commands inside Claude Code:
/plugin marketplace add epilot-dev/agent-toolkit-for-epilot
/plugin install epilot-core@agent-toolkit-for-epilot
/reload-plugins
ChatGPT and Codex support plugins, including skills and MCP tools. Local and Git marketplaces are authoring and team-distribution sources; adding this repo to Codex does not publish it to a ChatGPT workspace or the universal directory. For desktop testing, use a personal or repository marketplace. For workspace distribution, use the workspace's supported import or publishing flow. Public distribution requires submission and review by OpenAI.
See OpenAI's plugin packaging guide for current installation and distribution steps. This repository is the source package; it does not imply a published ChatGPT listing.
Both MCP servers are hosted over HTTPS: the epilot MCP at
https://mcp.epilot.io/mcp and the Volt UI MCP at
https://volt-ui.epilot.io/api/mcp. Neither requires a local process, so they
work in sandboxed surfaces such as Claude Cowork and ChatGPT. Only the App
inspector needs a local Node.js environment.
Connect https://mcp.epilot.io/mcp in an MCP-capable client and authenticate
when prompted. Optionally add https://volt-ui.epilot.io/api/mcp for Volt UI
components and design tokens. Direct MCP access provides documentation, API
discovery, organization tools, and design-system lookups; it does not install
this package's skills. Available tools remain subject to the client's
capabilities and your epilot permissions.
When the epilot MCP is first used, your agent may ask you to authenticate and select an organization. Access remains subject to your epilot permissions.
Access level is chosen on the epilot login screen when the connection is
approved: read-only is preselected, and read-and-write is an explicit
choice. To change it later, re-authenticate the connection and choose again.
For setups that must never write, connect the server as
https://mcp.epilot.io/mcp?access=read — that URL is enforced read-only
regardless of the approved consent.
- Understand epilot entities, relations, journeys, workflows, automation, portals, Apps, and integrations.
- Decide whether a requirement belongs in native configuration, an App, or an external integration.
- Configure schemas, journeys, products and pricing, workflows, automations, portals, and permissions end to end.
- Build workflows with phases, conditional branches, retry loops, and automated steps from a compact process description.
- Scaffold and validate Apps with the current epilot CLI and manifest schema.
- Discover current documentation and exact OpenAPI operations through MCP.
- Inspect the connected organization's actual schemas and configuration when the task depends on live state.
- Design native-feeling App surfaces with current Volt UI components and tokens.
The toolkit keeps stable routing and implementation guidance in focused skills. Changing platform facts remain in their authoritative sources and are retrieved only when needed:
| Source | Purpose |
|---|---|
| Agent skills | Stable mental models, decision rules, and delivery workflows |
| epilot MCP | Current docs, OpenAPI discovery, schemas, and organization state |
| Volt UI MCP | Current components, props, and design tokens |
| Project files and CLI | The App's actual manifest, code, and validation behavior |
.
├── .agents/plugins/marketplace.json # Codex marketplace
├── .claude-plugin/marketplace.json # Claude Code marketplace
├── .github/ # CI and contribution templates
└── plugins/
└── epilot-core/
├── plugin.json # portable manifest + OpenAI listing metadata
├── mcp.json # portable MCP configuration
├── .codex-plugin/plugin.json # Codex manifest
├── .claude-plugin/plugin.json # Claude Code manifest
├── .mcp.json # Codex and Claude MCP configuration
└── skills/
├── epilot-platform-guide/
├── build-epilot-app/
├── integrate-with-epilot/
├── configure-epilot/
├── build-epilot-journey/
└── build-epilot-workflow/
Skills are maintained in English and instruct the agent to respond in the user's language. Customer-facing copy follows the requested audience or the existing resource's locale, including consistent German du/Sie usage. API identifiers, schema keys, enum values, and commands are not translated.
The plugin listing and friendly skill labels currently use English, while the
brand remains epilot. The portable manifest has no standard locale map;
do not add invented en/de objects to string fields or assume automatic
listing translation. Each skill's agents/openai.yaml supplies its display
name, short description, and starter prompt.
Before a release, run the behavioral checks:
claude plugin eval in Claude Code and the manual script in other clients.
Structural checks do not establish model behavior.
- A supported coding agent
- Node.js 22 or newer for local validation and the App inspector
- An epilot account for organization-specific MCP operations
Documentation and API discovery may be available without organization access; organization-specific inspection depends on authentication and permissions.
Node.js 22 or newer is required. Skill validation also needs Python 3.9+
and the pinned development dependency in requirements-dev.txt. These are
maintainer checks; installing the plugin does not require PyYAML.
The toolkit has no installed JavaScript package dependencies.
python3 -m pip install -r requirements-dev.txt
npm run check
npm run inspect:app -- /path/to/an/epilot-appThe app inspector is a fast local preflight. The epilot CLI remains authoritative for platform validation:
npx @epilot/cli app validate
npx @epilot/cli app deploy --dry-runSkills contain decision guidance and compact source maps, not copied platform manuals. For changing behavior, prefer current epilot docs through MCP or https://docs.epilot.io/, current CLI help, manifest schemas, and the executable project itself. App scaffolds intentionally do not copy toolkit skills into generated repositories.
See CONTRIBUTING.md before opening a pull request and SUPPORT.md for the appropriate support channel. Report security issues privately according to SECURITY.md.
Licensed under the Apache License 2.0. See NOTICE for attribution and trademark information.