Benefits costs your assistant can back up.

evco's public connector gives Claude, ChatGPT and Grok seven tools for estimating what medical, dental and vision benefits cost, drawn from real premiums across evco's book of business. Every answer arrives with its range, its basis and its assumptions.

MCP endpointhttps://mcp.evcoinsurance.com/mcp

No sign-in. Agents can read the full tool list as JSON at /tools.

What comes back with every answer

This call ran against evco's model when this page was built. Every estimate carries the same parts, so your assistant can show its work.

You ask
What would medical benefits cost for a 30-person company in Colorado?

Plain words. No forms, no sign-in.

Your assistant calls
estimate_benefits_cost({"headcount": 30, "employer_state": "CO"})

A named tool with typed inputs, so the number comes from evco's data, not the model's memory.

evco returns a range
$898 to $1,476per enrolled employee per month, medical. Middle estimate $1,151.

The middle half of likely outcomes, not one figure that sounds more certain than it is.

With its basis
44 companies and 748 enrolled employees, data as of 2026-08.

Your assistant can say how much evidence sits behind the answer.

And what it assumed
  • Industry: book average (not given)
  • Participation: 0.64 (peer median)
  • Age: not applied (census date of birth pending)

Anything you didn't give is filled in and named, so you know what to add to narrow the range.

Every time
Estimate, not a quote. Based on evco's book of business and the information you shared; actual premiums depend on carrier underwriting.

An estimate is never presented as a quote.

Seven tools, in the order you'd use them

Read straight from the server, so this list always matches what your assistant sees.

  1. 1

    Estimate

    Get a cost range for your company and see what drives it.

    • Estimate benefits cost

      Reads only
      estimate_benefits_cost

      Returns monthly premium and employer cost ranges from evco's model, with support, confidence, assumptions, and data month. Requires request.headcount. Give the company's state as request.employer_state (two-letter code, e.g. "CO") or request.locations; optional age bands, household mix, contributions, industry, or a payroll connection. Estimate only, not an insurance quote.

    • Explain an estimate

      Reads only
      explain_estimate

      Summarizes the drivers, support, confidence, assumptions, and data date in a prior estimate response. Pass the complete response; this tool produces no new price.

  2. 2

    Compare

    Place a premium you already pay against companies like yours.

    • Compare to peer estimates

      Reads only
      compare_to_peers

      Returns the model-predicted premium range for a headcount or size band, optionally filtered by line, industry, and state, plus the position of a supplied premium. Thin segments use a broader prediction with lower confidence.

    • Explain a published benchmark

      Reads only
      explain_benchmark

      Returns one published, privacy-floored benchmark cell for a line and optional size band, industry, and state, including its quartiles and data month. Cells below the release floor are not returned.

  3. 3

    Refine with payroll

    Optional. Connect payroll so the estimate uses your real headcount and ages.

    • Connect payroll

      Acts on request
      connect_payroll

      Starts a payroll connection with a supported provider. Returns a connect URL and short-lived opaque connection reference; does not return employee rows. Optionally pass the requester's email (and name, company) only if they asked evco to follow up; it creates a sales lead when payroll connects.

      Starts a connection through Finch. Creates a follow-up lead with evco only if you ask for one.

    • Check payroll connection

      Reads only
      connection_status

      Returns the status and provider for a short-lived connection reference. Does not return employer or employee details.

    • Disconnect payroll

      Deletes
      disconnect_payroll

      Revokes the payroll connection identified by its opaque reference and deletes that reference.

What the tools will not do

These limits are built into the server, not left to your assistant's judgment.

  • No employee records

    Payroll connections keep headcount, age bands, state and 3-digit ZIP mix. Names, pay and individual rows never reach your assistant or evco's model.

  • No small groups

    A figure is published only when at least 10 companies stand behind it, so no single employer can be picked out.

  • No changes you didn't ask for

    Five tools only read. Connecting and disconnecting payroll are the only actions, and each runs on your request.

  • No quotes

    Every answer is an estimate. Actual premiums depend on carrier underwriting.

evco logs which tool ran, when, how long it took, the client app, and coarse inputs such as headcount band and state. Never your conversation. Disconnect payroll any time with the disconnect tool. Full policy: AI connectors and payroll connections.

Add it to your assistant

Directory listings are in review. Until then, add evco as a custom connector.

  • Claude

    Settings, then Connectors, then Add custom connector. Paste the MCP endpoint.

  • ChatGPT

    Settings, then Apps and Connectors. With developer mode on, create a connector with the MCP endpoint.

  • Grok

    Add a custom connector at grok.com/connectors with the MCP endpoint.

MCP endpointhttps://mcp.evcoinsurance.com/mcp

Setup guide: evcoinsurance.com/connector. Questions: marco@evcoinsurance.com.