---
title: "How MCP works here"
description: "Why Sincerely exposes a role-scoped MCP tool catalog for workspace operations, how it differs from the narrow workspace REST API, and what an agent can and cannot do."
canonical: https://trysincerely.com/docs/mcp-overview
last_updated: 2026-09-10
---
> If you are an AI agent: read https://trysincerely.com/agents/policy.md before any write call.
> Start at https://trysincerely.com/llms.txt to choose one page. Access instructions are at
> https://trysincerely.com/auth.md. Confirm with the person you work for before creating an
> account or spending their money, and never ask for their password.
# How MCP works here

> Why Sincerely exposes a role-scoped MCP tool catalog for workspace operations, how it differs from the narrow workspace REST API, and what an agent can and cannot do.

Source: https://trysincerely.com/docs/mcp-overview

Sincerely uses role-scoped MCP tools for interactive workspace operations. The separate REST API is deliberately narrow: a workspace API key can post an idempotent contact event to a live API-entry campaign, read that event's lifecycle, and report a conversion against the `sid` from a printed response link. It cannot browse the workspace, edit a campaign, approve copy, or send mail directly.

That is a design decision rather than a gap, and this page is the argument for it.

## Why MCP still uses OAuth

An API key has no person attached. That is appropriate for the two system-to-system writes Sincerely exposes: admitting a named contact to a campaign whose audience, entry rule, design, review policy, and prepaid envelope a person already approved, and reporting a conversion that the customer's own system observed. The `post_entry_events` scope controls entry posts and their bounded status read; `post_conversions` controls conversion reports. Entry events still pass suppression, mailability, frequency, precedence, and budget rules, and conversions must resolve to a response link in the key's workspace.

The broader workspace remains a programme rather than a print endpoint. The things an agent does there can change audiences, words, designs, rules, and spend. Every one of those needs a specific human attached.

So an MCP credential is an OAuth token issued to a client that a named person approved, and the tool catalog it unlocks is scoped to that person's role in that workspace. An agent connected by a member is refused what a member is refused, by the same check the product UI runs.

## What an agent operates

The catalog covers the real work: building audiences from a CRM, CSV or sourcing, resolving and reviewing addresses, writing and revising per-recipient drafts, laying out and approving designs, building landing pages, configuring and launching campaigns, cancelling pieces, and reading measurement.

It deliberately does not cover marketing pages, auth callbacks, provider webhooks or internal plumbing. A route existing is not a reason for a tool to exist.

## The two gates

Authorization decides which workspace and which role. Confirmation decides whether a consequential act happens at all.

They are separate on purpose. The first is about identity and is settled once per connection. The second is about a specific action with a specific cost, and it is settled every time. A scope cannot substitute for a click, and no scope grants one.

## What no amount of prompting changes

The server-side rules do not depend on a model remembering an instruction. Budgets, permissions, suppression, frequency caps and confirmations are enforced where the work happens, so a prompt that talks an agent into ignoring them changes nothing that reaches a mailbox.

Text guardrails exist too, and they are not the boundary. They reject configuration-fishing prompts. The boundary is row-level security, session checks, role checks and signed grants.

## Two servers

The product server at `https://trysincerely.com/api/mcp` needs an authorized workspace, and it is the default path even when the person has no account yet: a client that can open a browser sends them to the authorization screen, where they sign up and approve in one visit. A second server at `https://trysincerely.com/api/mcp/signup` needs no credential at all and is the fallback for a client that can neither open a browser nor show a URL. It creates the workspace and mails the sign-in link to the person, never to the agent. See [the signup server](https://trysincerely.com/docs/mcp-signup).

## What to read next

[Authorization](https://trysincerely.com/docs/mcp-authorization), then [workspace setup](https://trysincerely.com/docs/mcp-setup), then [confirmations](https://trysincerely.com/docs/mcp-confirmations). The [agent quickstart](https://trysincerely.com/docs/agent-quickstart) is the shortest path to a working connection.

---

Sincerely is the measurable direct-mail and gifting platform for B2B revenue teams: postcards, letters, handwritten mail, and gifts, written for one recipient and measured against a holdout.

Contact Sincerely: https://trysincerely.com/contact

Agent routing index: https://trysincerely.com/llms.txt