Skip to main content
Agent Toolkit for epilot Beta

Give your AI assistant the context to work with epilot

Focused workflows and live platform tools for configuring epilot, creating customer journeys, and building Apps and integrations. Works in Claude, ChatGPT, Codex, and any MCP client.

View installationView on GitHub
Built on Agent Plugins 1.0Skills + MCPOpen source ready

The Agent Toolkit and the epilot MCP server are available in Beta on all plans. Your use is subject to your epilot agreement and epilot's terms for beta features; tools can change, see the changelog.

CodexChatGPTClaude
Agent Toolkit for epilotSkills route the workOnly the relevant guidance is loaded
epilot MCPDocs, APIs, org state
Volt UI MCPComponents and tokens
Use cases

Describe the outcome. The agent does the epilot part.

Everyday tasks from real epilot projects — phrased the way you would ask for them, executed against current APIs and your organization's actual configuration.

Build Apps

  • “Add a grid-connection tab to the opportunity page”
  • “Build a custom journey block with IBAN validation”
  • “Show live meter readings as a customer portal widget”
  • “Call an external API from a flow action, keeping credentials server-side”

Build integrations

  • “Sync ERP orders into epilot entities with stable unique IDs”
  • “Update the JSONata mapping for inbound meter readings”
  • “Receive and verify epilot webhooks in an external service”
  • “Import a small dataset of contacts without writing a script”

Configure epilot

  • “Extend the contact schema after checking what depends on it”
  • “Create products and prices for a new tariff”
  • “Wire a journey submission to an automation that starts a workflow”
  • “Find every journey and automation that uses an email template”
Two parts

Plugin or MCP server? Usually both.

The plugin carries the know-how. The MCP server carries the live facts. Install the plugin where your client supports it; connect the MCP server everywhere else.

epilot plugin

epilot-core

Skills that teach the agent how to work with epilot: which architecture to choose, how to build Apps and integrations, how to configure the platform, and how to design native UI. Includes the MCP configuration.

Use it when

  • The agent will build or configure something
  • You want epilot’s proven patterns, not improvisation
  • You work in Claude Code, Codex, or ChatGPT
Needs a client that supports Agent Plugins. Installing it also connects the MCP server.

epilot MCP server

mcp.epilot.io

Hosted tools that return what is true right now: how an organization is configured, what depends on what, the current entity schemas, every published API operation, and the documentation.

Use it when

  • You only need the live tools, or your client offers connectors but no plugins (Cursor, VS Code, custom agents)
  • You ask questions about a live organization
  • You build your own agent and need tools, not prompts
Works in any MCP client. Connect with ?access=read for an enforced read-only connection.

Plain code against the public APIs needs neither. Use the SDK or CLI, and point any model at docs.epilot.io/llms.txt. Read the full comparison.

Why it matters

Platform expertise without a giant static prompt

The toolkit keeps stable decision guidance close to the agent and retrieves changing facts from their authoritative source when they are needed.

Start with the right architecture

Route a requirement to native configuration, an epilot App, or an external integration before writing code.

Build against current contracts

Discover epilot documentation, OpenAPI operations, entity schemas, and UI components instead of guessing from stale examples.

Understand the connected organization

Use the epilot MCP to inspect the organization's actual schemas and configuration when the task depends on live state.

Included workflows

One toolkit, five focused skills

Each skill stays focused on one workflow, so the agent loads only the epilot guidance that is relevant to the current task.

01

Understand epilot

Explains entities, relations, journeys, workflows, Apps, and the boundary between configuration and code.

02

Build an App

Scaffolds, extends, validates, and troubleshoots Apps with the current CLI, manifest schema, and App Bridge.

03

Connect a System

Designs reliable inbound, outbound, batch, webhook, and bidirectional connections without requiring an App.

04

Configure epilot

Sets up schemas, journeys, products, pricing, workflows, automations, portals, and permissions end to end.

05

Build a Journey

Creates and updates customer-facing forms and funnels, including steps, logic, copy, and design.

MCP capabilities

What the epilot MCP server can do

17 tools, built for the person configuring epilot. Curated tools only where a workflow spans several APIs or needs validation first; one generic route for every other published operation.

Configuration graph

How an organization is set up, what is connected to what, and what breaks if something changes. Also the way to list journeys, webhooks, portals, designs, and schemas.

  • search_configuration
  • get_config_dependencies
  • get_config_impact

Journeys

Read a journey as summary or editable definition, then create or update it with its entity mapping in one call. Step wiring is validated before anything reaches the API.

  • get_journey
  • create_journey
  • update_journey
  • get_journey_mapping
  • update_journey_mapping

Workflows and entity model

Compile a compact graph of tasks, branches, and loops into a valid workflow with linked automations, with dry runs before writing. Read compact tenant schemas that would not fit raw.

  • create_workflow
  • update_workflow
  • get_entity_schema

Any epilot API

Discover, describe, and execute every published OpenAPI operation, entity search included. Secrets are redacted; writes need mcp:write.

  • search_api_operations
  • describe_api_operation
  • call_api_operation

Documentation and connection

Search and read docs.epilot.io, and check who you are connected as.

  • search_docs
  • fetch_doc
  • whoami
Plain: works with mcp:readHighlighted: requires mcp:writeRuns as you, in your organization, PII-masked by default
Open the MCP server reference
Get started

Choose the setup for your client.

Claude, ChatGPT, Codex, and every other Agent Plugins client install the full plugin with skills. Any MCP client can also connect the hosted MCP server directly.

  • Each user connects their own epilot account
  • Read-only is preselected; write access is an explicit choice
  • Available actions follow your epilot permissions
  • Only connect to the official endpoint mcp.epilot.io

Full setup guide · Security best practices

Claude.ai, Claude Desktop, and Claude Cowork

Install the full plugin under Customize → Plugins once your workspace admin has made it available. Or connect the MCP server as a custom connector:

  1. Open Settings → Connectors → Add custom connector.
  2. Name it epilot, enter the URL below, and click Connect.
  3. Sign in to epilot, choose the organization, and choose read-only or read-and-write access.
https://mcp.epilot.io/mcp

Claude Code · full plugin with skills

/plugin marketplace add epilot-dev/agent-toolkit-for-epilot
/plugin install epilot-core@agent-toolkit-for-epilot
/reload-plugins
Claude Code · MCP server only
claude mcp add --transport http epilot https://mcp.epilot.io/mcp

For CI or headless use, reference an epilot API token from .mcp.json instead of OAuth. See the setup guide.

Start with an epilot App

Use the current CLI scaffold, then let the toolkit guide implementation, validation, and native UI design.

Open the quick start