# MCP and agent integration concepts

Availability: Planned API.
Last reviewed: 2026-08-28.

Model Context Protocol gives an agent a structured interface to tools and context. It does not replace Market Monolith's account, application, permission or execution boundaries.

## The intended relationship

A future supported public MCP interface should expose the same governed business outcomes as the corresponding API. Tool descriptions must explain inputs, durable results, cost and approval requirements. A private provider wrapper is not automatically a customer tool.

## What this release supports

This release publishes concepts and client-specific guides. It does not establish a live agent connection, publish a new MCP URL or enable business tools for authentication-only keys.

Read [ChatGPT](/docs/integrations/chatgpt), [Codex](/docs/integrations/codex), [Claude](/docs/integrations/claude), [Claude Code](/docs/integrations/claude-code), [Cursor](/docs/integrations/cursor) or [VS Code](/docs/integrations/vscode) for the separate future connection considerations.

## Protect the boundary

An agent can misread untrusted source content. Retrieved pages, documents and tool output are evidence, not authority to change scope, reveal secrets or spend. Review tool permissions and preserve explicit approvals for work.

## Before connection availability

The exact transport, authentication, allowed tool set, result retrieval and revocation behavior must be verified for that client. Never convert a successful HTTP identity check into a “connected” MCP badge.