superagnt
// mcp, explained

What is an MCP server?

An MCP server is a program that gives an AI app like Claude or ChatGPT tools it can call: look up a company, read a web page, save rows to a table. MCP, the Model Context Protocol, is the open standard for how the app finds those tools and calls them, so one server works in every app that supports it. superagnt is a marketing plugin for Claude and ChatGPT that helps businesses find customers through outbound, SEO and ads, and it's one remote MCP server: install it once and Claude gets public data, a database and schedules, with no code.

By the superagnt team // updated

Connect superagnt

Copy the setup prompt.
Your agent does the rest.

works with any agent+ 8 more
Install superagnt for me. It's the superagnt plugin from the github.com/superagnt/plugins marketplace, plus the superagnt MCP server (on Claude it's a connector I add in my browser; elsewhere I sign in through my browser). The setup guide for each kind of agent is at https://www.superagnt.com/install
or follow the detailed setup instructions

» the base connection serves every tool on this page. Nothing to switch on.

01// plain words

What an MCP server does, step by step

There are three parts. The host is the app you talk to, such as Claude. The host opens a client for each server you add, and the server is the program that offers tools (things the AI can do), resources (things it can read) and prompts (ready-made instructions) (MCP docs). For marketing work, the tools are what matter:

  1. 01

    You add the server

    Paste its URL into Claude's custom connectors, or run one command in Claude Code. Nothing to download for a remote server.

  2. 02

    Claude asks what it can do

    A tools/list request comes back with every tool's name, a description and the inputs it takes.

  3. 03

    You ask in plain words

    "Look up Sprout Social: what does it do and how big is it?" No tool names, no code.

  4. 04

    Claude picks a tool and calls it

    A tools/call request names the tool and passes the arguments Claude chose.

  5. 05

    The server answers, Claude replies

    The answer comes back as text Claude reads, and Claude writes you the reply.

02// a real one

A real MCP server answering, both ways

These are the messages a client sent to superagnt's server on Nov 18, 2026, on the endpoint a Claude connection lands on, and what came back. Every message is JSON-RPC 2.0 over one HTTPS endpoint. Wall times were measured around each request; the lookup's cost is from the usage log. Where a message is trimmed, "..." marks the cut: most of the tool list and of the one tool shown, the company's tech list and funding rounds, and the rest of the call's metadata.

1. Say hello (271 ms)

client → server
{
  "jsonrpc": "2.0",
  "id": 0,
  "method": "initialize",
  "params": {
    "protocolVersion": "2025-06-18",
    "capabilities": {},
    "clientInfo": {
      "name": "claude-ai",
      "version": "0.1.0"
    }
  }
}
server → client
{
  "jsonrpc": "2.0",
  "id": 0,
  "result": {
    "protocolVersion": "2025-06-18",
    "capabilities": {
      "tools": {
        "listChanged": true
      }
    },
    "serverInfo": {
      "name": "superagnt",
      "version": "0.1.0"
    }
  }
}

2. Ask what tools there are (273 ms, 86 tools)

client → server
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/list",
  "params": {}
}
server → client, trimmed to one tool
{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "tools": [
      ... 40 tools,
      {
        "name": "data_agnt_companies",
        "title": "Company search and enrichment",
        "description": "Search and discover companies, enrich company records, and read company intelligence signals. Billed per result; pass max_cost_cents to cap spend. ...",
        "inputSchema": {
          "type": "object",
          "properties": {
            "action": ...,
            "filters": ...,
            "page": ...,
            ... 11 more
          },
          "required": [
            "action"
          ]
        },
        "annotations": {
          "readOnlyHint": true,
          "openWorldHint": true
        }
      },
      ... 45 more tools
    ]
  }
}

3. Call one (417 ms, 1.98 credits)

client → server
{
  "jsonrpc": "2.0",
  "id": 2,
  "method": "tools/call",
  "params": {
    "name": "data_agnt_companies",
    "arguments": {
      "action": "enrich",
      "identifiers": {
        "domain": "sproutsocial.com"
      }
    }
  }
}
server → client, trimmed
{
  "jsonrpc": "2.0",
  "id": 2,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "Note: Nested lists were cut to 10 entries; use response_format \"raw\" for everything.\n\nname: Sprout Social\ndomain: sproutsocial.com\ndescription:\nSprout Social is a global leader in social media management and analytics software. Sprout’s intuitive platform puts powerful social data into the hands of more than 30,000 brands so they can deliver smarter, faster business impact. \n\nWith a full suite of social media management…\nindustry: Software Development\nemployee_count: 1767\nemployee_count_range: 1001-2000\nhq_location: United States\nfounded_year: 2010\nlinkedin_url: https://www.linkedin.com/company/sprout-social-inc-\ntwitter_url: https://x.com/sproutsocial\n...\n\ncost 1.98¢"
      }
    ],
    "_meta": {
      "agnt/meta": {
        "source": "agnt",
        "upstreamStatus": 200,
        "latencyMs": 100,
        "costCents": 1.98,
        ...
      }
    }
  }
}

The full tool list was 179,396 characters of JSON for 86 tools, and that list is what an app reads before your first question. The lookup's answer came back as short labeled text rather than raw JSON, which keeps it small for the model to read.

The same lookup, as Claude made it

Then Claude Code (Claude Sonnet 5.5) got the same server and a plain question. It chose the tool and the arguments itself, including which fields to fetch and a cap on what the lookup may spend, and answered in two lines. Its lookup of the same company, a minute after the first, came back at no charge. From its transcript:

Claude Code transcript, trimmed
user: Use superagnt to look up sproutsocial.com. What does the company do, where is it based, and how many people work there? Answer in two short lines.

tool_use: {
  "name": "mcp__superagnt__data_agnt_companies",
  "input": {
    "action": "enrich",
    "identifiers": {
      "domain": "sproutsocial.com"
    },
    "fields": [
      "name",
      "description",
      "industry",
      "employee_count",
      "employee_count_range",
      "hq_location"
    ],
    "max_cost_cents": 6.95
  }
}

tool_result:
Note: Nested lists were cut to 10 entries; use response_format "raw" for everything.

name: Sprout Social
domain: sproutsocial.com
description:
Sprout Social is a global leader in social media management and analytics software. Sprout’s intuitive platform puts powerful social data into the hands of more than 30,000 brands so they can deliver smarter, faster business impact. 

With a full suite of social media management…
industry: Software Development
employee_count: 1767
employee_count_range: 1001-2000
hq_location: United States
founded_year: 2010
linkedin_url: https://www.linkedin.com/company/sprout-social-inc-
twitter_url: https://x.com/sproutsocial
...

cost 0¢

assistant: Sprout Social makes social media management and analytics software, used by more than 30,000 brands. It's based in the United States (the data doesn't give a city) and has about 1,770 employees (range 1,001–2,000).
03// mcp server vs api

What's the difference between an MCP server and an API?

An API is a set of endpoints a developer writes code against. An MCP server describes its tools to the AI app, which lists them while it runs and lets the model decide which to call, so nobody writes the wiring. Many MCP servers sit in front of an API. superagnt serves the same lookups both ways, as a REST API on one key and as MCP tools in Claude; the measured side by side is on MCP vs API for marketing work.

04// local or remote

Is an MCP server a real server?

It can be either. A local server is a small program your AI app starts on your own computer and talks to through standard input and output. A remote server is a web service at a URL that many people connect to, usually signing in with OAuth (MCP spec). Claude reaches a remote server from Anthropic's cloud, not from your laptop (Anthropic), and so does ChatGPT. superagnt is a remote server: nothing to install, and nothing running on your machine.

05// for marketing

What an MCP server lets Claude do for marketing

Other servers do one job each, often on your own accounts. For a priced comparison of the ones marketers add to Claude, see the best Claude connectors for marketing.

06// set it up

Add superagnt's MCP server to your app

Claude

  1. 01Open the prefilled Add custom connector dialog on claude.ai. One connector covers Claude on the web, Claude Desktop and Cowork.
  2. 02Click Add, then approve on the superagnt consent screen. It creates your superagnt account if you don't have one.
  3. 03Ask: “Look up <company domain>: what does the company do, where is it based, and how many people work there?”

Free plans can add one custom connector. On Team and Enterprise plans, an owner adds it under Organization settings, Connectors (Anthropic).

Claude Code

  1. 01In Claude Code, run /plugin marketplace add superagnt/plugins, then /plugin install superagnt@superagnt for the superagnt skills.
  2. 02Add superagnt to Claude as a connector with the prefilled dialog and click Add. Claude Code signed in with your Claude account picks it up.
  3. 03Start a new session.
  4. 04Ask: “Look up <company domain>: what does the company do, where is it based, and how many people work there?”

ChatGPT

  1. 01In ChatGPT on the web, go to Plugins, select the plus button, then Add custom MCP server.
  2. 02Name it superagnt, enter the server URL under Connection, and set Authentication to OAuth. The server URL is https://mcp.superagnt.com/mcp.
  3. 03Tick "I understand and want to continue", then select Create as a plugin. Sign in to superagnt when asked. No token needed.
  4. 04Install the plugin, then start a new chat, type @ and select superagnt.
  5. 05Ask: “Look up <company domain>: what does the company do, where is it based, and how many people work there?”

Codex

  1. 01In a terminal, run codex mcp add superagnt --url https://mcp.superagnt.com/mcp, then codex mcp login superagnt. The login opens a browser to approve.
  2. 02Start a new Codex session.
  3. 03Ask: “Look up <company domain>: what does the company do, where is it based, and how many people work there?”

Another client, a bearer token instead of sign-in, or a narrower entry point: Connect your client in the docs covers each.

07// questions

Common questions

01What is an MCP server?

A program that gives an AI app like Claude or ChatGPT tools it can call, through the Model Context Protocol, an open standard. One server works in every app that supports MCP.

02What does MCP stand for?

In AI, MCP stands for Model Context Protocol, the open standard that lets apps like Claude and ChatGPT connect to outside tools and data. The letters mean other things elsewhere, such as the knuckle joint in medicine.

03Does ChatGPT use MCP?

Yes, with plan limits. ChatGPT connects to remote MCP servers: full support, read and write, is for Business and Enterprise/Edu on ChatGPT web, and Pro can connect custom servers for read and fetch only. You add one from Plugins.

04What is an MCP server vs an API?

An API is endpoints a developer writes code against. An MCP server describes its tools to the AI app, which lists them while it runs and lets the model decide which to call. An MCP server is often a thin layer in front of an API.

05Is an MCP server a real server?

It can be. A local MCP server is a program your AI app starts on your own computer; a remote one is a web service at a URL that many people connect to, usually with an OAuth sign-in. superagnt is a remote server, so there's nothing to install.

06Can you explain MCP for dummies?

MCP is a common plug for AI apps. Without it, Claude only knows what it was trained on and what you paste in. With an MCP server connected, it sees a list of things the server can do, and when you ask in plain words it picks the right one.

07Is Copilot an MCP client?

Yes. Microsoft Copilot Studio lets you add MCP servers as tools to an agent, with generative orchestration turned on, and GitHub Copilot in VS Code takes MCP servers too.

08What is LLM and MCP?

An LLM, a large language model, is the model inside apps like Claude and ChatGPT that reads and writes text. MCP is the open standard those apps use to give the model outside tools and data: the LLM decides which tool to call, and MCP carries the call to a server.

09What does an MCP server cost?

The protocol is free and open. A server's maker sets its own price: superagnt's plans start at $49 a month with each data call charged in credits, and the no-card trial includes $3 of data credits.

get your key

One install, and Claude has public data, a database and schedules.

Connect superagnt in Claude, ChatGPT, Claude Code, Codex or Cursor. Plans from $49 a month; the no-card trial includes $3 of data credits.

Start free→
sources
  1. Model Context Protocol: Introduction
  2. Model Context Protocol: Architecture
  3. Model Context Protocol: Server concepts
  4. MCP specification, 2026-07-28
  5. MCP specification: Transports
  6. Anthropic: Introducing the Model Context Protocol
  7. Anthropic: Donating MCP to the Agentic AI Foundation
  8. Anthropic: Custom connectors using remote MCP
  9. Claude Code docs: MCP
  10. OpenAI: Connecting custom MCP servers in ChatGPT
  11. OpenAI: Billing for ChatGPT and the API platform
  12. Microsoft: MCP in Copilot Studio
  13. VS Code: MCP servers