MCP setup and management

Beta — Real Talk Studio’s MCP server and platform connectors are in beta. Capabilities, URLs, and setup steps may change; we recommend testing in a non-production agent before wide rollout.

What is MCP?

Model Context Protocol (MCP) is an open standard for connecting AI assistants and agents to external systems — tools, APIs, and data — in a consistent way. Instead of each platform inventing its own integration pattern, MCP defines how a client (your copilot or agent runtime) talks to an MCP server that exposes tools the model can call when a user’s request needs real actions (for example, creating a simulation or fetching structured data). Real Talk Studio provides an MCP server so compatible platforms can discover our tools (such as creating roleplay simulations) and invoke them when the conversation warrants it, without custom code in your agent for every capability.

How it works

Registration — In your platform’s admin or agent settings, you add Real Talk Studio as an MCP server using the MCP server URL (typically https://<your-host>/api/mcp). No studioId query parameter is required when using OAuth.
OAuth discovery — Compatible clients discover authentication via protected resource metadata (/.well-known/oauth-protected-resource/api/mcp, plus host-root compatibility) and same-origin authorization server metadata (/.well-known/oauth-authorization-server). There are no shared org API keys.
Consent — An organization admin completes a browser consent screen, selects the organization (if they belong to more than one), and approves access. Tokens are user-delegated and org-scoped.
Discovery — The MCP server publishes a tool catalog: names, descriptions, and parameters. Your platform’s orchestration layer uses those descriptions to decide when a tool is relevant to the user’s message.
Invocation — When the model chooses a tool, the client sends a structured request with a Bearer access token. The server runs the action and returns a result (for example, a link to a simulation).
Transport — Real Talk Studio’s MCP server uses Streamable HTTP, the transport pattern supported by modern MCP clients. No extra plugins are required beyond what the platform’s MCP onboarding asks for.
Updates — When we add or change tools on our side, connected clients typically refresh the catalog automatically, so new capabilities can appear without you redeploying custom integrations.

Org admins can review and revoke MCP authorizations under Organization → Integrations.

Tool catalog (org-scoped)

Catalog, analytics, and iteration tools require OAuth (organization admin consent). They operate on the connected organization only.

Tool	Purpose
`create_simulation`	Create a 1:1 roleplay in the org catalog; returns a `/roleplay/custom/{slug}` preview link (login + org credits required).
`create_team_simulation`	Create a team (2–4 AI) meeting roleplay; same preview link pattern.
`list_simulations`	List active (non-archived) org simulations (default 20, max 50).
`get_simulation`	Read scenario content (learning objective, skill good/poor examples, context, structure, personas). Works for archived rows too.
`archive_simulation`	Soft-archive a throwaway simulation (hides from org catalog). Does not archive personas or remove existing studio links.
`add_roleplay_to_studio`	Link a simulation to a studio. If the studio has `allowed_domains`, returns an iframe snippet plus domains (not a bare clickable URL). Rejects archived roleplays.
`get_org_report`	Compact usage summary for the last N days (default 30). Counts studio-attributed completions only — preview-only sessions without a studio are excluded.
`test_simulation`	Sync-test a 1:1 or team scenario: predicted example transcript + realism score/issues/recommendations. Not a live engine run (no mood/length-band/split-engine coverage; no policy/product-knowledge files).
`refine_simulation`	Propose (`apply: false`, default) or confirm (`apply: true` + `expected_updated_at`) content refinements from plain-language feedback. Covers backstory, structure, avatar fields, learning objective, and skills (name, description, good/poor examples — the score rubric). 1:1 only; rejects archived. Not wholesale Enhance.

Iteration loop (1:1)

Typical workflow after create:

test_simulation → predicted transcript + realism analysis (read the disclaimer — this is not a live engine run).
refine_simulation with feedback and apply: false → proposed recommendations + expected_updated_at.
Confirm with the user, then refine_simulation with apply: true, the chosen recommendations, and the same expected_updated_at.
test_simulation again to verify.
Optionally add_roleplay_to_studio.

You may pass realism recommendations from test_simulation straight into refine_simulation with apply: true (no propose call). If the scenario was edited in the Real Talk UI meanwhile, confirm fails with a conflict — re-propose and retry with the new expected_updated_at. Limits: Team scenarios can be tested (kind: "team") but not refined via MCP. Archived scenarios can be tested (read) but not refined until restored in the org catalog UI. Typical catalog workflow: create → review via preview link → optionally add_roleplay_to_studio → list_simulations / get_simulation to browse → archive_simulation for throwaways → get_org_report for usage.

Connect a platform

Use the guides below to wire Real Talk Studio into each supported environment. More platforms will be added over time.

Microsoft Copilot Studio

Add Real Talk Studio as an MCP tool and test simulations from the Copilot test panel.

Claude

Add Real Talk Studio as a custom remote MCP connector via Customize → Connectors (web, Desktop, and Claude Code).

ChatGPT

Add Real Talk Studio as a custom remote MCP app in ChatGPT developer mode (web).

Slack

Use Real Talk Studio from Slack via Marketplace AI apps (e.g. Claude) and the provider’s MCP connector settings.

What you’ll need in general

A Real Talk Studio MCP server URL for your deployment (for example https://www.realtalkstudio.com/api/mcp).
OAuth — most clients register themselves automatically. Keep a pre-registered Client ID as fallback if a platform cannot do dynamic registration.
An organization admin Real Talk Studio account to complete consent.
Appropriate admin or customizer access in the target platform to add tools and connections.
Network access to reach our MCP endpoint over HTTPS (see Firewall configuration if traffic is restricted).

For product support, credentials, or environment-specific questions, see How to get help.

Overview

Product

Guides