LoyJoy

LoyJoy Headless

Tell Claude what your AI Agent should do.

LoyJoy is a platform for customer service AI agents that you can control entirely from your AI assistant through an MCP server. Describe the task, the knowledge and the process in conversation. LoyJoy applies the change in your tenant, you approve the release. The same works with ChatGPT, Microsoft Copilot and other MCP-capable AI assistants. Included in all plans.

Request a demoRegister for a webinar

In 60 seconds

Building an agent has never been this easy

One sentence to your assistant, a finished draft in staging, an analysis in conversation. Your assistant does the work, you approve.

Illustrative example with sample data. The video is served from loyjoy.com, no data is sent to third parties.

How it works

You say it. Claude builds it.

No click paths, no forms, no switching tools. Claude sets up process, knowledge and integrations, you approve.

1

Describe

You tell your assistant what the agent should handle: requests, knowledge sources, process steps, tone of voice. The assistant translates that into the structure of your LoyJoy tenant.

2

Review

The change is created in staging. You inspect process, instruction and knowledge sources in the Manager, or have them explained in conversation, before anything goes live.

3

Publish

Approval is a separate step and follows your roles and permissions. Whatever is published is versioned and traceable at any time.

An example

What a change sounds like in conversation

Claude, connected to your LoyJoy tenant

Build me an agent for questions about premium adjustments. Use our customer letters as the knowledge source, contract data comes from the policy administration system.

Created in staging: agent Premium adjustment with three process steps, knowledge search across your customer letters and an API client on the policy system. The draft is ready for approval.

  • 1. Identify the contract from the customer number
  • 2. Explain the adjustment based on the customer letter
  • 3. Offer and document a suitable tariff change

Open the draft in the Manager and publish

How many requests of this kind did we get last quarter?

Illustrative example. Which tools the assistant may use is governed by your roles and permissions.

Build, analyse, improve

What your team does in the AI assistant

Once the tenant is connected, you work with agents, processes and analytics in natural language.

Create agents.
Set up new AI Agents from a description, including process steps and knowledge sources.
Edit agents.
Rewrite instructions, add modules and sharpen answers without leaving your assistant.
Query analytics.
Compare KPIs per process and period and generate your own analyses as a table or infographic.
Check knowledge.
Search the knowledge base semantically and find gaps before customers run into them.
Derive improvements.
Have drop-offs and knowledge gaps analysed and apply the suggestions straight to the draft.
Stay in control.
Roles, permissions, versioning and approval stay in LoyJoy, and every call is traceable.

Live: analytics in Claude

From a question to a finished analysis, with no data export and no prebuilt view in the backend.

LoyJoy analytics in Claude via MCP

Analytics in Claude, in 2 minutes

Click to play. The video is only loaded from YouTube after the click (privacy friendly).

Setup in 2 minutes

How to connect LoyJoy Headless

Shown for the Claude desktop app. The guide for your assistant follows right below.

Adding the LoyJoy MCP server as a custom connector in Claude
  1. 1

    Add the MCP server

    Add a custom connector with the URL https://app-cloud.loyjoy.com/mcp and sign in with your LoyJoy account. No API key needed.

  2. 2

    Add the plugin

    Add the marketplace loyjoy/loyjoy-skills in the plugin settings of your assistant and install the LoyJoy plugins.

  3. 3

    Start modelling

    Describe your change, for example: “Set the opening hours of the agent Live Chat to 10am to 6pm.” The assistant works in staging, you test and publish.

Guide for your assistant

Click your assistant to expand the steps.

Claude (web and desktop)
  1. Open Settings › Connectors and add a custom connector named “LoyJoy Manager” with the URL https://app-cloud.loyjoy.com/mcp. Sign in to LoyJoy when prompted.
  2. Open Settings › Plugins, click Add › Add marketplace, enter loyjoy/loyjoy-skills and sync.
  3. In the Discover tab, add the plugins LoyJoy Headless and LoyJoy Phone Agent Builder.

Claude Team and Enterprise: only an owner can add custom connectors. The owner adds the connector once under Organization settings › Connectors, then each member signs in with their own LoyJoy account.

ChatGPT (Codex)
  1. Open Plugins and click Add › Add MCP server. Name “LoyJoy Manager”, type Streamable HTTP, URL https://app-cloud.loyjoy.com/mcp. Leave all other fields empty.
  2. Save and sign in to LoyJoy.
  3. Click Add › Add plugin marketplace: source loyjoy/loyjoy-skills, Git ref main.
  4. Under Personal, click + next to loyjoy-headless and loyjoy-phone-agent-builder.
Claude Code
/plugin marketplace add loyjoy/loyjoy-skills
/plugin install loyjoy-headless@loyjoy-skills
/plugin install loyjoy-phone-agent-builder@loyjoy-skills
Microsoft Copilot, OpenWork and other assistants
  1. The assistant must support remote MCP servers with sign-in (OAuth).
  2. Add https://app-cloud.loyjoy.com/mcp as an MCP server and sign in to LoyJoy.
  3. If the assistant supports skills but not plugins, copy the folders plugins/*/skills/ from the GitHub repository loyjoy-skills into its skills directory.

For every assistant: no API key needed. Your role in the LoyJoy tenant determines what the assistant may change.

The architecture behind it

LoyJoy remains your execution layer

Your assistant handles the interaction. LoyJoy manages agents, knowledge, processes, versions, roles, permissions and analytics. That keeps the choice of an employee assistant independent from the platform running your customer processes.

Understand the three MCP directions

What you need

  1. 1. A LoyJoy tenant. LoyJoy Headless is included in all plans.
  2. 2. An MCP-capable assistant. Claude, ChatGPT, Microsoft Copilot, OpenWork or any other MCP-capable assistant. In Claude Team and Enterprise, an owner adds the connector once for the whole organization.
  3. 3. Permissions. The assistant only uses the tools approved for your access.
  4. 4. LoyJoy Skills. The free plugin for your assistant, open on GitHub: loyjoy/loyjoy-skills

Add the marketplace (Claude Code)

/plugin marketplace add loyjoy/loyjoy-skills/plugin install loyjoy-headless@loyjoy-skills

In the Claude desktop app, add the marketplace loyjoy/loyjoy-skills and install the plugin in the Discover tab. All steps for every assistant are in the step-by-step guide.

The AI assistant itself may be subject to separate licences and data processing terms from its provider.

Webinar

See LoyJoy Headless live

In our free webinars we show how an agent is created in conversation, reviewed and published. Dates and registration are on the webinar page.

See the dates

Frequently asked questions about LoyJoy Headless

See LoyJoy Headless in action

We show you, on your own use case, how your team builds, reviews and publishes AI Agents from Claude, ChatGPT, Microsoft Copilot or other MCP-capable AI assistants.

Request a demo