MCP server for AI assistants

What this is#

Couplr Business Tools is a small MCP server with calculators for HVAC, plumbing, and electrical business owners. You give it your own numbers; it works out a labor rate, checks a job’s profit, models what booking more of your leads would be worth, prices a maintenance plan, or finds the break-even for a new hire or truck.

MCP (Model Context Protocol) is a standard way for an AI assistant, such as ChatGPT or Claude, to use outside tools. Once you connect this server, your assistant can run these calculators for you during a conversation and explain the results.

• No Couplr account is needed. There is no sign-in, API key, or signup.
• It only calculates. It cannot read or change anything in a Couplr account.
• It uses only the numbers you provide. It does not supply market prices, wage benchmarks, or promised results.

What it can and cannot do#

Before you connect#

• An AI assistant or other MCP client that can connect to a remote MCP server by URL without authentication. See “Connect a client”.
• Your own business numbers. Each calculator lists exactly what it needs in the “Tool reference”. If you leave something out, the assistant should ask you for it rather than guess.

Do not enter customer names, addresses, contact details, passwords, or payment details. The calculators accept only the numeric fields, trade, and currency listed in the tool reference.

Quickstart#

1. Open your MCP client’s settings for adding a custom or remote MCP server. Steps for specific clients are in “Connect a client”.
2. Add this server URL: https://uoax74kzmplxwtkqxvaqb5edzm0xurio.lambda-url.us-east-1.on.aws/mcp. Choose no authentication.
3. Start a new conversation and make sure the Couplr tools are enabled for it.
4. Ask a question using your own numbers. For example: “My HVAC shop has 2 techs at $30/hour plus 25% burden, 40 paid hours a week for 50 weeks, 50% billable utilization, and $50,000 annual overhead. What hourly labor rate gives a 20% operating margin? Use USD.”
5. Review the answer and the assumptions listed with it. Change any input and ask again to compare scenarios.

For that example the tool returns a break-even rate of 100 USD and a target rate of 125 USD per billable hour. These figures are an example computed from the sample inputs, not a recommendation.

Connect a client#

The server does not offer the older HTTP+SSE transport or a GET event stream: GET and DELETE on the URL return 405. It is meant to be called by an MCP client, not from a web page: requests that carry a browser Origin header are refused with 403.

The steps below follow each vendor’s own documentation as checked on 2026-11-14. Menu names can change; the linked vendor pages are the authority.

ChatGPT#

Use ChatGPT on the web. Whether you can add a custom MCP server depends on your plan and, in a workspace, on your administrator’s settings.

1. Go to ChatGPT Plugins at chatgpt.com/plugins.
2. Select the plus button, then Add custom MCP server.
3. Enter a name, for example “Couplr Business Tools”, and optionally a description.
4. Under Connection, enter the Server URL: https://uoax74kzmplxwtkqxvaqb5edzm0xurio.lambda-url.us-east-1.on.aws/mcp.
5. For authentication, choose No authentication.
6. Review the risk warning and select I understand and want to continue.
7. Select Create as a plugin, then review the tools ChatGPT discovered.
8. Install the plugin you just created.
9. Start a new conversation, type @ in the prompt box, select the plugin, and ask your question.

ChatGPT supports the MCP Apps standard, so results can appear as an interactive card with editable inputs.

Source: OpenAI: Add custom MCP server, OpenAI: Connect and test, OpenAI: Plugins quickstart.

Claude#

Custom connectors are available on the Free, Pro, Max, Ultra, Team, and Enterprise plans; the Free plan allows one custom connector. On Team and Enterprise plans an Owner adds the connector under Organization settings > Connectors, and members then connect to it.

1. In Claude on the web or desktop, go to Customize > Connectors.
2. Select Add, then Add custom connector.
3. Enter a name and the remote MCP server URL, https://uoax74kzmplxwtkqxvaqb5edzm0xurio.lambda-url.us-east-1.on.aws/mcp, then select Continue.
4. For authentication, choose No sign-in.
5. Select Add. If the connector shows Connect, select it.
6. In a conversation, select the + button, choose Connectors, and turn the connector on. Then ask your question.

Claude asks before running a tool; choose Allow once or Always allow. Connectors added on the web or desktop are also available in the Claude mobile apps.

Source: Claude: Add a custom connector, Claude Help: custom connectors.

Claude Code#

Add the server

claude mcp add --transport http couplr-business-tools https://uoax74kzmplxwtkqxvaqb5edzm0xurio.lambda-url.us-east-1.on.aws/mcp

Check that it connected

claude mcp list

To share the server with everyone working in a project, add --scope project, which writes this entry to .mcp.json:

.mcp.json

{
  "mcpServers": {
    "couplr-business-tools": {
      "type": "http",
      "url": "https://uoax74kzmplxwtkqxvaqb5edzm0xurio.lambda-url.us-east-1.on.aws/mcp"
    }
  }
}

Remove the server

claude mcp remove couplr-business-tools

Source: Claude Code: MCP.

MCP Inspector (for testing)#

The MCP Inspector is the protocol project’s own test tool. It needs Node.js 22.19.0 or newer. Use it to see the tools and try a call without an AI assistant.

Open the Inspector connected to the server

npx @modelcontextprotocol/inspector --server-url https://uoax74kzmplxwtkqxvaqb5edzm0xurio.lambda-url.us-east-1.on.aws/mcp --transport http

Or list the tools from the command line

npx @modelcontextprotocol/inspector --cli https://uoax74kzmplxwtkqxvaqb5edzm0xurio.lambda-url.us-east-1.on.aws/mcp --transport http --method tools/list

Source: MCP Inspector.

Tool reference#

The server has 5 tools: plan_labor_rate, analyze_job_profit, plan_lead_recovery, price_maintenance_plan, plan_hire_break_even. All of them are read-only calculations. Live at the endpoint today: plan_labor_rate, analyze_job_profit, plan_lead_recovery.

plan_labor_rate#

Plan a sustainable labor rate. Model the hourly service labor rate needed to cover paid technician time, payroll burden, and annual overhead at a user-chosen operating margin. For HVAC, plumbing, and electrical business owners comparing billable utilization. Returns a cost breakdown and utilization scenarios. Ask for missing inputs; do not invent wages, overhead, utilization, or industry benchmarks. This is a standalone planning model, not a price-book update.

All fields are required. Fields not listed here are rejected. Numbers must be JSON numbers, not strings.

Example

Request: “My HVAC shop has 2 techs at $30/hour plus 25% burden, 40 paid hours a week for 50 weeks, 50% billable utilization, and $50,000 annual overhead. What hourly labor rate gives a 20% operating margin? Use USD.”

Arguments for plan_labor_rate

{
  "trade": "hvac",
  "currency": "USD",
  "technicians": 2,
  "hourly_wage": 30,
  "payroll_burden_pct": 25,
  "paid_hours_per_week": 40,
  "paid_weeks_per_year": 50,
  "billable_utilization_pct": 50,
  "annual_overhead": 50000,
  "target_operating_margin_pct": 20
}

Example result, key figures (computed by the tool from the arguments above):

• Target rate / billable hour: 125 USD
• Break-even rate / billable hour: 100 USD
• Annual billable hours: 2,000
• Annual labor revenue needed: 250,000 USD

analyze_job_profit#

Check a job’s profitability. Analyze a hypothetical HVAC, plumbing, or electrical job from selling price, materials, technician-hours, loaded wage, other direct costs, and allocated overhead. Compare the current price with break-even and a user-chosen target margin. Does not create, edit, send, or charge an invoice or estimate. Ask for missing cost inputs; never assume a market price.

All fields are required. Fields not listed here are rejected. Numbers must be JSON numbers, not strings.

Example

Request: “For a plumbing job in USD: $1,000 price, $200 materials, 8 total tech-hours at $40 loaded cost, $80 other direct costs, and $100 overhead. Check a 30% target margin.”

Arguments for analyze_job_profit

{
  "trade": "plumbing",
  "currency": "USD",
  "selling_price": 1000,
  "materials_cost": 200,
  "technician_hours": 8,
  "loaded_hourly_wage": 40,
  "other_direct_costs": 80,
  "allocated_overhead": 100,
  "target_margin_pct": 30
}

Example result, key figures (computed by the tool from the arguments above):

• Profit after entered costs: 300 USD
• Margin after entered costs: 30%
• Break-even price: 700 USD
• Price at target margin: 1,000 USD

plan_lead_recovery#

Plan missed-lead recovery. Compare current and target booking rates for an HVAC, plumbing, or electrical service business. Model additional booked jobs, revenue, and contribution after variable job costs, capped by available additional monthly job capacity. Uses only the user’s assumptions, not promised conversion improvements. Provides practical follow-up steps. Does not contact leads or access customer records.

All fields are required. Fields not listed here are rejected. Numbers must be JSON numbers, not strings.

Example

Request: “My electrical shop gets 100 leads a month, books 30%, and wants to test 50%. Average revenue is $500 USD, contribution margin is 40%, and we can handle only 10 more jobs.”

Arguments for plan_lead_recovery

{
  "trade": "electrical",
  "currency": "USD",
  "monthly_leads": 100,
  "current_booking_rate_pct": 30,
  "target_booking_rate_pct": 50,
  "average_job_revenue": 500,
  "contribution_margin_pct": 40,
  "additional_monthly_capacity": 10
}

Example result, key figures (computed by the tool from the arguments above):

• Additional jobs / month: 10
• Additional revenue / month: 5,000 USD
• Additional contribution / month: 2,000 USD
• Annualized additional revenue: 60,000 USD

price_maintenance_plan#

Price a maintenance plan. Model the monthly and annual price of a recurring maintenance plan (service agreement or membership) for an HVAC, plumbing, or electrical business from the user’s own cost to deliver it: visits per year, technician-hours and materials per visit, the yearly cost of member perks, per-plan overhead, and a user-chosen margin. Returns break-even and target prices and margin scenarios. Ask for missing inputs; do not invent visit times, costs, or typical plan prices. Does not create, change, or sell a plan in any system.

Not live yet: This tool ships with server version 1.1.0. The public endpoint currently runs 1.0.0, so your assistant will not see it yet.

All fields are required. Fields not listed here are rejected. Numbers must be JSON numbers, not strings.

Example

Request: “Price an HVAC maintenance plan in USD: 2 visits a year, 1.5 tech-hours per visit at $40 loaded cost, $15 materials per visit, $20 a year in member perks, $10 a year overhead per plan, and a 40% margin.”

Arguments for price_maintenance_plan

{
  "trade": "hvac",
  "currency": "USD",
  "visits_per_year": 2,
  "technician_hours_per_visit": 1.5,
  "loaded_hourly_wage": 40,
  "materials_cost_per_visit": 15,
  "member_perks_cost_per_year": 20,
  "annual_overhead_per_plan": 10,
  "target_margin_pct": 40
}

Example result, key figures (computed by the tool from the arguments above):

• Monthly price at target margin: 25 USD
• Annual price at target margin: 300 USD
• Break-even monthly price: 15 USD
• Annual cost to deliver one plan: 180 USD

plan_hire_break_even#

Find the break-even for a new hire or truck. Model how many additional jobs an HVAC, plumbing, or electrical business needs each month to cover the yearly cost of adding a technician, a vehicle, or both, and how long expected work would take to recover one-time costs. Uses the user’s own wages, vehicle costs, average job revenue, and contribution margin. Ask for missing inputs; do not invent wages, vehicle costs, job volumes, or benchmarks. This is not a forecast of demand and not hiring, tax, or lending advice.

Not live yet: This tool ships with server version 1.1.0. The public endpoint currently runs 1.0.0, so your assistant will not see it yet.

All fields are required. Fields not listed here are rejected. Numbers must be JSON numbers, not strings.

Example

Request: “My plumbing shop is adding a tech at $60,000 a year plus 25% burden, with a truck costing $12,000 a year, $3,000 in other yearly costs, and $9,000 up front. Jobs average $500 USD at a 50% contribution margin, and I expect 36 more jobs a month. How many jobs do I need to break even?”

Arguments for plan_hire_break_even

{
  "trade": "plumbing",
  "currency": "USD",
  "annual_wages": 60000,
  "payroll_burden_pct": 25,
  "annual_vehicle_cost": 12000,
  "other_annual_costs": 3000,
  "one_time_costs": 9000,
  "average_job_revenue": 500,
  "contribution_margin_pct": 50,
  "expected_jobs_per_month": 36
}

Example result, key figures (computed by the tool from the arguments above):

• Break-even jobs / month: 30
• Break-even revenue / month: 15,000 USD
• Monthly contribution after added costs: 1,500 USD
• Months to recover one-time costs: 6

Result shape#

Every tool returns the same structure, both as structured data and as the same JSON in a text block, so hosts without a card still get the complete result.

format is one of money, percent, number, or text. Money and percentages are rounded to two decimal places.

Example: full result of plan_labor_rate for the quickstart request

{
  "kind": "plan_labor_rate",
  "title": "Your labor-rate plan",
  "input": {
    "trade": "hvac",
    "currency": "USD",
    "technicians": 2,
    "hourly_wage": 30,
    "payroll_burden_pct": 25,
    "paid_hours_per_week": 40,
    "paid_weeks_per_year": 50,
    "billable_utilization_pct": 50,
    "annual_overhead": 50000,
    "target_operating_margin_pct": 20
  },
  "currency": "USD",
  "metrics": [
    {
      "label": "Target rate / billable hour",
      "value": 125,
      "format": "money"
    },
    {
      "label": "Break-even rate / billable hour",
      "value": 100,
      "format": "money"
    },
    {
      "label": "Annual billable hours",
      "value": 2000,
      "format": "number"
    },
    {
      "label": "Annual labor revenue needed",
      "value": 250000,
      "format": "money"
    }
  ],
  "columns": [
    "Billable utilization",
    "Billable hours / year",
    "Break-even rate",
    "Target hourly rate"
  ],
  "rows": [
    [
      {
        "label": "Utilization",
        "value": 40,
        "format": "percent"
      },
      {
        "label": "Hours",
        "value": 1600,
        "format": "number"
      },
      {
        "label": "Break even",
        "value": 125,
        "format": "money"
      },
      {
        "label": "Target",
        "value": 156.25,
        "format": "money"
      }
    ],
    [
      {
        "label": "Utilization",
        "value": 50,
        "format": "percent"
      },
      {
        "label": "Hours",
        "value": 2000,
        "format": "number"
      },
      {
        "label": "Break even",
        "value": 100,
        "format": "money"
      },
      {
        "label": "Target",
        "value": 125,
        "format": "money"
      }
    ],
    [
      {
        "label": "Utilization",
        "value": 60,
        "format": "percent"
      },
      {
        "label": "Hours",
        "value": 2400,
        "format": "number"
      },
      {
        "label": "Break even",
        "value": 83.33,
        "format": "money"
      },
      {
        "label": "Target",
        "value": 104.17,
        "format": "money"
      }
    ]
  ],
  "assumptions": [
    "Annual base wages: 120000 USD; payroll burden: 30000 USD; overhead: 50000 USD.",
    "Break-even rate = (annual wages + payroll burden + allocated overhead) / annual billable technician hours.",
    "Target rate = break-even rate / (1 − target margin). Margin is a share of revenue, not a markup on cost.",
    "Each hour is one technician-hour. Two technicians on one job for an hour consume two billable hours.",
    "All paid hours incur the entered average wage. Incorporate overtime premiums into that average if applicable.",
    "This labor-only model assumes labor revenue covers all entered costs. Do not double-count costs or overhead recovered elsewhere. No market-rate or demand forecast is included."
  ],
  "nextSteps": [
    "Compare your actual billed technician-hours with paid hours for the last four weeks.",
    "Check that owner pay, vehicles, insurance, office staff, and other overhead are allocated once.",
    "Compare the target rate with your current realized labor revenue per billed hour before changing prices."
  ]
}

Errors#

A failed tool call comes back as a normal tool result marked as an error, with a message the assistant can read and act on. No partial result is returned.

Read-only behavior and retries#

• Every tool is read-only. Each is annotated readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false. Nothing is created, changed, sent, or charged.
• There are no mutations, so the server has no approval or confirmation step of its own. Your MCP client may still ask your permission before it runs a tool; that prompt comes from the client and its settings, not from this server.
• Retrying is safe. The same inputs always produce the same result, and the server keeps no state between requests.
• Results are planning arithmetic. They are hypothetical figures from your assumptions, not customer estimates, invoices, forecasts, or financial advice.

Privacy and security#

• The server receives only what your assistant sends in a tool call: the numeric assumptions, trade, and currency. The input schemas accept nothing else.
• Results are computed in memory and returned. The application does not write inputs or results to storage, does not use cookies or analytics, and does not log calculator inputs or results.
• The service does not connect to a Couplr account, contact customers or leads, or send email or text messages.
• Your AI assistant’s provider handles your conversation, including what you type and the results, under its own terms and privacy policy.
• The interactive card loads no external scripts, styles, or fonts.
• The card includes an optional link to couplrfsm.com. If you follow it, the Couplr website can remember the referral for 30 days in a first-party cookie; the card has a checkbox to turn that off. Calculator inputs are never included in the link.

Because the endpoint is unauthenticated, treat it like any public calculator: do not send personal or confidential information in tool inputs.

Troubleshooting#

Permissions: there are none to manage. The server has no user accounts, so there is nothing to grant or revoke. To stop using it, remove the server from your MCP client; see “Disconnecting”.

Disconnecting#

Remove or disable the server in your MCP client’s settings. Nothing is stored about you on the server, so there is no account to close and nothing to delete on Couplr’s side.

Versions and compatibility#

Clients that require a newer MCP protocol version than those listed cannot connect until the server is updated. Tool names, field names, and the result shape are stable within server version 1.x.

Related: Model Context Protocol, Couplr.