MCP Agent Use Cases: 8 AI Agents You Can Build by Role

📖 TL;DR

• An MCP agent is four things: a model, two or three MCP servers, a system prompt, and a guardrail on writes.
• Sales → HubSpot or Salesforce + Slack + Calendar. Marketing → Ahrefs or Semrush + Exa + Notion.
• Social media → X and TikTok have official ad MCP servers. Organic posting still needs Zapier or your own server.
• Support → Stripe + Slack + Linear. Data → PostHog or Amplitude + your warehouse. Engineering → GitHub + Sentry + Linear.
• Three servers is the ceiling for most agents. Past that, tool-selection accuracy falls off a cliff.
• Model choice is a cost decision, not a quality one. Haiku for routing, Sonnet for reasoning, Opus for long chains.

Every team I talk to wants the same thing from MCP. Not a protocol tour — an agent that does their actual job.

The sales lead wants call prep. The marketer wants a content brief that cites real keyword data. The support lead wants refund triage.

These are all the same build. Only the servers change.

So this post is organised the way you actually think about it: by role. For each one I list the MCP servers worth connecting, the hosted URLs, a model recommendation, and example prompts that work on day one.

I have kept every URL genuine. Where a service has no official hosted MCP server, I say so instead of inventing one.

Anatomy of an MCP Agent (Four Parts, No More)

Strip away the frameworks and every MCP agent is the same four things.

The fourth row is the one people skip. An agent with a Stripe token can issue refunds. An agent with a CRM token can overwrite deal stages.

I put one line in every system prompt I write: propose writes, never execute them, until I say go.

How to Pick MCP Servers for Any Agent

The instinct is to connect everything. Resist it.

Every connected server injects its full tool schema into the model's context on every single request. A large server like GitHub runs to roughly 18,000 tokens on its own.

Connect six of those and you have burned your context window before the user types a word. Worse, tool-selection accuracy drops as near-identical tool descriptions pile up.

I use a three-server rule, and each server plays a distinct part:

1. System of record — where the truth lives. HubSpot, GitHub, Stripe, your warehouse.
2. Context source — what fills the gaps. Exa for the web, Notion for internal docs, Calendar for timing.
3. Delivery channel — where the output lands. Slack, Notion, Linear.

Record, context, channel. Almost every useful agent below is that shape.

1. The Sales Agent: CRM Hygiene and Call Prep

Sales reps lose hours to two chores. Researching a prospect before a call, and updating the CRM after it.

Both are pure context assembly. That is exactly what an MCP agent is good at.

My recommended build: Calendar + HubSpot + Exa. The agent reads tomorrow's meetings, pulls each account's CRM history, and researches recent news.

Model: Claude Sonnet 5. Call prep involves synthesis across three sources, which is where cheaper models start dropping details.

Prompts that work immediately:

• Prep me for tomorrow's calls. For each one, pull the deal stage, last contact date, and anything newsworthy about the company.
• Which deals have had no activity in 21 days and are still marked as open?
• Draft a follow-up email for the Acme deal referencing what we discussed last time.

Watch the write scope. A HubSpot token with write access lets the agent change deal stages. Start read-only, and add writes once you trust its judgement.

2. The Marketing Agent: Briefs Backed by Real Data

Most AI content workflows fail for one reason. The model has no idea what people actually search for.

It guesses at keywords, invents volumes, and produces a brief that reads well and ranks nowhere.

Connecting a real SEO data source fixes that in one step.

My recommended build: Ahrefs + Exa + Notion. Keyword truth, live SERP context, and a place to file the output.

Model: Sonnet 5, or a stronger reasoning model when the brief spans a dozen competitor pages.

Prompts worth stealing:

• Find five keywords we rank on page two for, with difficulty under 30, and outline a post for the best one.
• Compare the top five results for this keyword and list every heading they cover that we do not.
• Which of our published posts have lost positions in the last 90 days?

---

Before you hand an SEO server real API credits, connect it in the browser and read its tool list. Ahrefs and Semrush both meter by call, and a chatty agent burns quota fast.

• Test any MCP server free →

3. The Social Media Agent: Where MCP Helps and Where It Does Not

This is the role with the biggest gap between expectation and reality, so let me be blunt about it.

Paid social has excellent official MCP coverage. Organic posting mostly does not.

X and TikTok both ship real hosted MCP servers, and both are aimed at advertising. There is no official hosted MCP server for scheduling an Instagram carousel or a LinkedIn post.

Zapier is the honest answer for scheduling. You pick the actions you want exposed, Zapier generates a private MCP endpoint, and the agent calls those actions as tools.

It is not elegant. It works today, and it covers the platforms nobody else does.

Model: Sonnet-class for ad analysis. A cheap model is plenty if the agent only drafts copy and reads metrics.

Prompts to start with:

• Which three ad sets had the worst ROAS last week, and what do the winning ones have in common?
• Pull our last 20 posts and tell me which format gets the most saves.
• Draft five hooks for this week's launch, each under 200 characters, in our usual voice.

4. The Customer Support Agent: Triage Before a Human Reads It

Support tickets arrive with no context. The agent's job is to attach it before a human opens the thread.

Who is this customer? What plan are they on? Did they just get charged twice? Is this a known bug?

Three servers answer all four questions.

• Stripe — https://mcp.stripe.com/ for subscription state, invoices and failed payments.
• Slack — https://mcp.slack.com/mcp to read the support channel and post the summary back.
• Linear — https://mcp.linear.app/sse to check whether the bug is already filed.

Model: Claude Haiku 4.5. Triage is high-volume and low-ambiguity, which is the cheapest model's sweet spot.

This is also the role where read-only really matters. A support agent should never issue a refund on its own initiative.

⚠️ Prompt injection is a live risk here. Ticket text is untrusted input written by strangers. An agent that reads tickets and holds a Stripe write token is one crafted message away from a bad day. Keep the refund tool out of its reach.

• Scan your MCP server →

5. The Data and Analytics Agent: Answers Without a Ticket

Every analytics team has the same queue. Twenty people asking questions that are one SQL query away from an answer.

An MCP agent with warehouse access clears most of that queue, as long as you keep it read-only.

My recommended build: one analytics source, one warehouse, Slack for delivery.

Model: a strong reasoning model. SQL generation against an unfamiliar schema is where the cheaper tiers start guessing at column names.

Non-negotiable: connect with a read-only role. Not a role you promise to use carefully — one the database will not let write.

6. The Engineering Agent: The Best-Served Role in MCP

Engineering has the deepest MCP coverage of any function. Almost every developer tool shipped a server first.

Three builds cover most of what engineering teams ask for:

• Engineering lead — GitHub + Linear + Slack. Sprint status without the standup.
• Full-stack shipper — GitHub + Vercel + Sentry. Ship, watch, roll back.
• Post-incident review — Sentry + Linear + Slack. A timeline written while it is still fresh.

Model: the top tier. Code reasoning across a diff is the one place a frontier model consistently earns its cost.

Only the GitHub server is free to connect with no key. Context7 is the other one, which makes the pair a good first test.

7. The Project Ops Agent: Meetings In, Tasks Out

Project managers spend their week converting one format into another. Meeting notes into tickets. Tickets into status updates.

That is mechanical work, and it is the easiest agent on this list to get right.

• Asana — https://mcp.asana.com/mcp
• Monday.com — https://mcp.monday.com/mcp
• Jira and Confluence — https://mcp.atlassian.com/v1/mcp
• Airtable — https://mcp.airtable.com/mcp
• Google Calendar — https://calendarmcp.googleapis.com/mcp/v1

My recommended build: Calendar + Notion + Slack. It reads yesterday's meetings, finds the notes, and drafts the action items.

Model: Haiku 4.5. Summarising and restructuring text does not need a reasoning model, and this agent runs daily.

8. The Finance and RevOps Agent: Revenue Questions, Answered Live

Finance questions are usually simple and always urgent. Which subscriptions failed to renew this week?

Payment platforms have solid MCP coverage, so this one is quick to stand up.

• Stripe — https://mcp.stripe.com/ for subscriptions, invoices, disputes and payouts.
• PayPal — https://mcp.paypal.com/mcp for orders and refunds.
• Cashfree — https://mcp.cashfree.com/mcp for India-first payment flows.

Join Stripe to Linear and Slack, and a spike in failed payments becomes a tracked issue rather than a Slack message nobody actions.

Model: Sonnet-class. Money questions deserve a model that checks its arithmetic against the tool output rather than guessing.

Which Model for Which Agent?

Teams overthink this. Model choice is a cost decision far more often than a quality one.

A triage agent reading tickets does not need frontier reasoning. A code-review agent across a 900-line diff does.

Start one tier below what you think you need. If the agent picks the wrong tool or drops a step, move up one and compare.

Test Before You Trust: Four Checks That Take Five Minutes

Here is the failure mode I see most. Someone wires four servers into an agent, it behaves strangely, and they blame the model.

Nine times out of ten the server was the problem. It exposed three tools instead of the twelve the docs promised, or its auth silently failed.

So before any URL above goes into an agent, run it through four checks:

1. Does it connect at all? Paste the URL into a tester and watch the handshake.
2. What tools does it really expose? Read the list. Compare it to what you assumed.
3. Does a real call return real data? A tools list that loads is not proof the tool works. Invoke one.
4. How big is the schema? A server with 40 verbose tools will crowd out everything else you connect.

---

This runs in the browser with no install. Paste any remote MCP URL, complete the OAuth flow or add a bearer token, and inspect the tool list and raw JSON-RPC responses before you commit.

• Test any MCP server free →
• Scan your MCP server →

Five Mistakes That Ruin Otherwise Good Agents

1. Connecting every server you can find. Three is the working ceiling. Each extra one costs context and accuracy.

2. Giving write access on day one. Run read-only for a week. Read the transcripts. Then decide which writes it has earned.

3. Using one generic system prompt for every agent. A sales agent and a support agent need different stop rules, not the same helpful-assistant boilerplate.

4. Skipping the token scope review. A GitHub token with repo scope reaches every private repository you can. Scope it down before it goes in.

5. Never evaluating the thing. If you cannot say whether last week's version was better, you are guessing.

Start With One Role, Not a Platform

The pattern repeats across every role here. One system of record, one context source, one delivery channel, and a model matched to the difficulty.

Sales gets a CRM and a calendar. Marketing gets keyword data. Support gets billing state. Engineering gets the repo and the error tracker.

Pick the role that loses the most hours this week and build that one. Test every server in the browser before you trust it with a token.

Related Guides

• What Is the Model Context Protocol (MCP)? A Developer's Guide
• AI Agent + MCP Explained: What Every Developer Needs to Know
• What Is an MCP Agent? Tool Calling Explained
• The Best AI Model for MCP Tool Calling
• MCP Context Bloat and Tool Search
• Build Your First MCP Agent With No Code
• What Is an MCP Eval, and Why You Need One

Further Reading

• Official: MCP Specification
• Official: MCP Server Registry
• GitHub MCP Server
• Stripe: Model Context Protocol docs
• Linear: MCP server docs
• Zapier MCP
• Anthropic: Building effective agents

---

Originally published on MCP Playground.