crm.care is now an MCP server

crm.care now speaks the Model Context Protocol. Install once in Claude Desktop, Cursor, Continue.dev, or any MCP-compatible client, and your AI agent can read crm.care's product info, search the blog, look up your campaigns and brand voice — all without leaving the chat.

There are two tiers: a public read-only server anyone can install (it answers questions about the product), and an authenticated per-workspace server for paying customers (it reads your campaigns + brand voice + budget, scoped to your workspace via a Personal Access Token).

This post walks through both.

Why MCP, why now

The AI agent ecosystem is splitting in two. On one side, agents that browse the public web — ChatGPT, Perplexity, Claude with web search — synthesise answers from whatever's indexed by Bing or Brave. On the other side, agents that work on behalf of a logged-in user — Claude Desktop, Cursor, Continue.dev — need authenticated access to that user's tools and data.

The Model Context Protocol is how the second category talks to those tools. Instead of every SaaS product building a custom Claude plugin, a custom Cursor integration, a custom Continue extension, MCP standardises the contract: tools, resources, prompts. Build one server, get every MCP-compatible client for free.

For a B2B SaaS like crm.care, this matters because the buyer's centre of gravity is shifting from the SaaS UI to the AI agent. Marketing operators on Account Engagement still spend most of their day inside AE and crm.care. But increasingly they also have a Claude window open — drafting copy, debugging, asking clarifying questions about their own data. If crm.care is invisible to that Claude window, we're invisible to the place the operator's attention is moving.

The MCP server fixes that.

The public tier — no auth required

The public MCP server lives at https://crm.care/api/mcp. It exposes seven read-only tools that answer questions about crm.care itself: what it does, who it's for, pricing, blog content, competitor comparisons, FAQ.

Install in Claude Desktop:

json
{
  "mcpServers": {
    "crm.care": {
      "command": "npx",
      "args": [
        "mcp-remote@latest",
        "https://crm.care/api/mcp"
      ]
    }
  }
}

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json (macOS; equivalent paths on Windows / Linux), restart Claude Desktop. Same config snippet works in Cursor and Continue.dev.

After installation, Claude has tools available like:

• crm_care_overview — what crm.care does
• crm_care_pricing — tiers and trial terms
• crm_care_icp — fit profile (and explicit non-fit signals)
• crm_care_search_blog(query) — search the blog
• crm_care_get_blog(slug) — fetch a specific post
• crm_care_compare(competitor) — vs Jasper / HubSpot / Marketo / Pardot
• crm_care_faq — canonical answers to common questions

The agent calls these automatically when the conversation warrants. Ask Claude "Is crm.care a fit for a Marketo shop?" and it'll call crm_care_icp, get our honest "no, we're AE-specific, here's what to look at instead" answer, and pass that on rather than guessing.

The authenticated tier — your workspace

The authenticated server lives at https://crm.care/api/mcp/v1. Same protocol, but with a Bearer-token Authorization header. The token authenticates to a specific (user, org) pair — your Claude can read your campaigns, never anyone else's.

Generate a token at /admin/api-keys inside your crm.care workspace. The token's shown once on creation alongside a pre-formatted config snippet — copy both, paste into your MCP client config, restart, done.

The current authenticated surface is read-only:

• list_campaigns — recent campaigns, filterable by status
• get_campaign(id) — full detail
• get_brand_voice — your configured voice
• get_budget_usage — this month's spend, quota usage
• search_campaigns(query) — keyword search
• list_recent_content — recent social posts and emails

Write tools — generate a campaign, publish to AE, edit brand voice — are deliberately out of scope for this first authenticated phase. We're proving the auth and tool plumbing with read-only traffic before exposing operations that burn budget or mutate AE state. Write tools land in a follow-up.

What you can ask your agent

A few example interactions, with the tool calls the agent makes under the hood:

"What campaigns am I running this week?" → list_campaigns(status: 'in_progress') → Claude summarises your active campaigns by objective and audience.

"How much AI budget have I burned this month?" → get_budget_usage() → Spend vs. cap, quota usage, days remaining.

"Find the post we wrote about Three Degrees of Attribution." → crm_care_search_blog(query: 'three degrees attribution') → returns the slug; Claude can then crm_care_get_blog(slug) for the full text and quote relevant sections.

"Help me draft a follow-up email for the Q3 launch campaign." → get_campaign(id: '...') to load the brief + brand voice → Claude drafts the email in your actual voice with context from the existing campaign plan.

The last one is the most interesting pattern. The agent isn't generating campaigns for you (that still happens in crm.care, where it lives in your campaign history). It's a research and drafting assistant that has access to the same workspace context you do — without you needing to copy and paste briefs into the chat.

What we're betting on

Two things shape the bet behind investing in MCP at this stage of the product.

One: marketing operators are early adopters of AI tooling, but late adopters of integrations between AI tooling. The path of least resistance has been to copy-paste between Claude and AE. MCP collapses that step. The first marketing-tools company to show up natively inside Claude wins the workflow.

Two: distribution matters. The MCP ecosystem in 2026 is where the App Store was in 2009 — small, technical, but compounding fast. Anthropic's official directory, mcp.so, smithery.ai, and the modelcontextprotocol/servers GitHub repo are real discovery channels for the AI-native developer audience. That audience overlaps significantly with the technically-inclined marketing ops persona we build for. Being visible there is a real channel, not an afterthought.

There's also a quieter benefit: dogfooding. Lee runs crm.care from inside Cursor most days. Having the MCP means we eat our own product before customers do — every UX friction surfaces immediately because we're the first ones hitting it.

Try it

The public tier is install-and-go: copy the config above, paste into Claude Desktop, restart, ask Claude "What does crm.care do?" — and watch it call our tools.

The authenticated tier needs a paid crm.care account (7-day free trial counts) → generate a token at /admin/api-keys → install with the pre-formatted snippet from the reveal dialog.

Full docs, both install paths, and copy-paste config snippets are at /mcp.

If you build something interesting with it — or hit something that doesn't work — drop us a note at hello@crm.care.