Overview

Regal MCP extends the same AI-assisted workflow pattern that powers Regal Copilot to any MCP-compatible client outside the Regal app. Teams can query Regal data, inspect workflows, and run guided actions from their own AI editors, internal tools, batch scripts, and custom applications — all scoped to their brand and gated by OIDC authentication.

MCP (Model Context Protocol) is the open standard that lets AI clients discover and call tools exposed by a server. Regal pairs its MCP tools with Skills — packaged workflow instructions that tell the AI how to use Regal tools for specific goals — so external users get guided workflows, not just raw API access.

When to use Regal MCP

Use case	What it enables
AI client integrations	Connect Cursor, Claude, ChatGPT, or your own AI client to Regal context and actions.
Batch analysis	Run structured analysis across calls, journeys, or contact data from scripts or notebooks.
Custom internal apps	Build internal tools that query Regal context and surface it to your team.
Workflow extensions	Extend Regal workflows into your own systems while keeping the AI grounded in Regal-specific Skills.

Access

Authentication

Regal MCP uses OIDC (OAuth 2.0) authentication via Okta. Every request to the MCP endpoint must include a valid Bearer token in the Authorization header. The token must contain claims for brand, sub, and roles — requests missing any of these are rejected before any tool executes.

Contact your Regal account team to obtain OIDC credentials.

MCP server configuration

The canonical config payload for Regal MCP is:

{
  "mcpServers": {
    "production": {
      "url": "https://mcp.regal.ai/v1/external-mcp/mcp"
    }
  }
}

Once you add this config to your AI client, you'll be prompted to log in to Regal via a browser the first time the MCP server connects. After authenticating, your session token is handled automatically — you don't need to hardcode any credentials.

Cursor

Open Cursor and go to Settings → Tools & MCP → Installed MCP Servers
Click Add MCP Server and paste the config below, or manually add it to .cursor/mcp.json in your project root (project-scoped) or ~/.cursor/mcp.json (global)
Fully quit and reopen Cursor — MCP servers only load at startup
On first use, Cursor will open a browser window prompting you to log in to Regal to complete authentication

{
  "mcpServers": {
    "production": {
      "url": "https://mcp.regal.ai/v1/external-mcp/mcp"
    }
  }
}

Note: Cursor has a limit of ~40 active tools across all MCP servers combined.

→ Cursor MCP docs

Claude Desktop

Open Claude Desktop and go to Settings → Developer → Edit Config. This opens (or creates) the config file in your default editor.
- macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
- Windows: %APPDATA%\Claude\claude_desktop_config.json
Add the Regal server entry inside the mcpServers object and save
Restart Claude Desktop
On first use, a browser window will open prompting you to log in to Regal to complete authentication

{
  "mcpServers": {
    "production": {
      "url": "https://mcp.regal.ai/v1/external-mcp/mcp"
    }
  }
}

→ Claude Desktop MCP docs

Claude Code

Run this command in your terminal to add Regal MCP:

claude mcp add --transport http --scope user production https://mcp.regal.ai/v1/external-mcp/mcp

--scope user makes the server available across all your Claude Code sessions. Use --scope local instead to limit it to the current project.
After running the command, verify it was added with claude mcp list.
On first use, a browser window will open prompting you to log in to Regal to complete authentication.

You can also configure it manually by editing ~/.claude/settings.json (user scope) or .claude/settings.json in your project root (project scope):

{
  "mcpServers": {
    "production": {
      "url": "https://mcp.regal.ai/v1/external-mcp/mcp",
      "transport": "http"
    }
  }
}

→ Claude Code MCP docs

ChatGPT

Full MCP connector support requires a ChatGPT Business or Enterprise workspace. Admins must first enable Developer Mode at Workspace Settings → Permissions & Roles → Connected Data.

Go to Settings → Connectors → Create
Enter a name (e.g. "Regal") and set the server URL to https://mcp.regal.ai/v1/external-mcp/mcp
Save — you'll be prompted to log in to Regal via browser to authorize the connection

→ ChatGPT MCP connector docs

Tools

All tools are scoped to the authenticated brand — no cross-tenant access is possible. Tools marked with * below currently require a UI session and may return an error on the external MCP endpoint; these are flagged as known gaps.

Read tools

Tool	Description
`list-ai-agents` / `list-agents`	List AI agents configured for your brand, with pagination and channel-type filtering.
`fetch-agent`	Fetch a specific agent by ID, including its full configuration and prompt.
`fetch-latest-agent`	Fetch the most recent version of an agent.
`list-ai-agent-variants`	List all variants for a given agent, including traffic allocation.
`fetch-test-cases`	Fetch test cases associated with an agent.
`explore-agent-context`	Returns workflow instructions for building and understanding AI agent configurations.
`summarize-agent-plan`	Returns a plan template to guide agent design decisions.
`generate-agent-config`	Returns section-by-section guidance for editing an agent config.
`get-prompting-best-practices`	Returns Regal's static prompting reference for agent builders.
`get-ai-agent-tool-context`	Returns the tool-context reference used when building agent tool configurations.
`fetch-simulation-transcript`	Fetch the transcript from a completed simulation run.
`list-recordings`	List call recordings, with filters for date range, agent, disposition, and phone number.
`fetch-recording-transcripts`	Fetch transcripts for one or more recordings (up to 5 SIDs per call).
`get-trackers-lookup`	Return the list of tracker IDs and names defined for your brand.
`list-analysis-data-points`	List analysis data point definitions used for call scoring and QA.
`list-journeys`	List journeys with search and pagination support.
`fetch-journey`	Fetch a full journey definition, including its flow and node configuration. Webhook auth headers are redacted.
`list-journey-versions`	List all saved versions of a journey.
`fetch-journey-version`	Fetch a specific historical version of a journey for diffing or review.
`list-profile-journey-executions`	List journey executions for a given contact profile.
`get-profile-journey-execution-nodes`	Get the node-level detail of a specific journey execution.
`lookup-profile`	Look up a contact profile by phone, email, name, or external ID. Returns multiple candidates if the query is ambiguous.
`fetch-profile`	Fetch a full contact profile by profile ID.
`fetch-profile-events`	Fetch event history for a contact profile, with type and cursor-based pagination.
`list-profile-versions`	List all saved versions of a contact profile.
`fetch-profile-version`	Fetch a specific historical snapshot of a profile.
`get-brand`	Fetch your brand's configuration, including telephony, compliance, and queue settings.
`list-users`	List agent and user accounts in your brand.
`fetch-user`	Fetch a specific user account by email, including queue assignments.
`list-queues`	List routing queues and their eligibility rules.
`list-tasks`	List tasks by state, queue, type, or phone number, with pagination.
`fetch-task`	Fetch a specific task by SID.
`get-task-events`	Fetch routing event history for a task.
`get-routing-rules`	Get routing rule definitions for your brand.
`get-routing-rule-versions`	Get version history for a named routing rule.
`query-regal-kb`	Query the Regal knowledge base with a natural language question.

Coming soon — Write tools

These tools are in development and will be available in a future release. They will require explicit human approval at the client level before execution.

Tool	Description
`create-agent`	Create a new AI agent.
`edit-agent`	Edit an existing agent (saves as a draft version).
`create-ai-agent-variant`	Create a new variant for an existing agent.
`set-variant-traffic-allocations`	Set traffic allocation percentages across all variants of an agent.
`create-test-cases`	Create test cases for an agent.
`edit-test-case`	Edit an existing test case.
`delete-test-cases`	Delete one or more test cases.
`start-simulations`	Start a simulation run against an agent and a set of test cases.
`get-simulation-progress`	Poll the progress of a running simulation.

Helpful Notes

Recommended starting prompts

These patterns work well with Regal MCP because they give the AI a bounded task, a clear Regal object, and an expected output.

Review AI agent transcripts

"Fetch the last 10 recordings for agent [name] and summarize the most common objections customers raised. Group by theme and flag any calls where the agent deviated from the expected flow."

Debug a task or routing issue

"Look up the task with SID [sid]. Fetch its routing events and tell me where in the routing flow it stalled or failed. Cross-reference the queue and routing rules if available."

Fetch and diff a journey definition

"Fetch journey [name or ID] and its two most recent versions. Summarize what changed between them, focusing on flow logic, node types, and any webhook configurations."

Inspect a contact's history

"Look up the profile for [phone or email]. Fetch their recent events and any active journey executions. Give me a timeline of their last 7 days of activity."

Analyze call quality for a segment

"List recordings from the last 7 days for disposition [value]. Fetch transcripts for up to 50 and identify the most common reasons customers were dissatisfied."

Review agent configuration before a launch

"Fetch the latest version of agent [name]. Summarize its prompt, configured tools, and any active variants. Flag anything that looks incomplete or inconsistent."

General approach

Start specific. Give the AI a named object (agent ID, recording SID, contact phone, journey name) rather than an open-ended question. Regal MCP tools return structured data — the more specific the input, the more useful the output.

Chain tools intentionally. Many useful workflows involve two or three tools in sequence: lookup-profile → fetch-profile-events → list-profile-journey-executions. Describe the end goal and let the AI plan the chain, or specify the steps yourself if you want predictable behavior.

Use Skills for repeatable work. If your team runs the same analysis or review workflow regularly, package it as a Regal Skill so the AI follows the same steps each time instead of improvising.

Keep humans in the loop on writes. Current write tools (when available) only save drafts — they don't publish or go live without a separate action. Still, confirm tool arguments before execution, especially in clients that show a preview before calling.

Watch for the routing rules gap. get-routing-rules and get-routing-rule-versions currently require a UI session and may return an error from the external MCP endpoint. This is a known issue — use list-queues as a workaround for queue-level context in the meantime.