Home

Connect an agent

Point your agent at Tallyho.

Tallyho is an MCP server with three tools: find, book and report. Add one URL to Claude or ChatGPT, and your agent can search the licensed plumbers near a job in San Francisco, Oakland, Berkeley or the Tri-Valley. It’s free for agents.

Live since 12 November 2026, and early. What is real today is spelled out below, before you rely on it.

For a real job

https://mcp.tallyho.so/mcp/v1

Use this one when someone needs a plumber. Searches, bookings and reports made here are kept, and reports move a real plumber’s numbers.

Only trying it out

https://mcp.tallyho.so/mcp/v1/test

The same tools and the same real plumbers. Nothing you do there counts: test searches, bookings and reports never change a likelihood.

01 · Set up

One URL, three clients.

Pick what the connection is for, then follow the steps for your client.

Snippets for

The snippets below use the live URL. Bookings and reports made through it count against real plumbers.

Claude Desktop

The same steps work on claude.ai. The connector belongs to your Claude account.

  1. Open Customize, then Connectors.
  2. Select Add, then Add custom connector.
  3. Name it Tallyho, paste the URL below as the server URL, and continue.
  4. Tallyho needs no authentication. If Claude asks, choose No sign-in. Then select Add.
  5. In a chat, open the + menu, choose Connectors, and check that Tallyho is on.
https://mcp.tallyho.so/mcp/v1

On a Team or Enterprise plan, an Owner adds the connector for the organization first. The Free plan allows one custom connector. Claude’s menus change, so Anthropic’s own instructions are the reference.

Claude Code

One command in a terminal.

  1. In a terminal, run the command below.
  2. Start Claude Code and run /mcp. tallyho should show as connected.
claude mcp add --transport http tallyho https://mcp.tallyho.so/mcp/v1

That adds Tallyho for you, in the project you ran it from. Add --scope user to have it in every project. To share it with everyone on a project instead, commit this as .mcp.json at the project’s root:

{
  "mcpServers": {
    "tallyho": {
      "type": "http",
      "url": "https://mcp.tallyho.so/mcp/v1"
    }
  }
}

Signed in to Claude Code with your Claude account? A connector you added in Claude is already listed under /mcp. Claude Code’s MCP documentation has the rest.

ChatGPT

A custom MCP server, by URL.

Add Tallyho to ChatGPT as a custom MCP server with the URL below. Tallyho needs no authentication, so choose none if ChatGPT asks.

https://mcp.tallyho.so/mcp/v1

This page doesn’t name ChatGPT’s menus: where the setting lives, and which plans have it, has changed more than once. OpenAI’s help page on custom MCP servers has the current steps.

Any other MCP client works the same way: Tallyho speaks Streamable HTTP at that URL, and asks for no key.

02 · First run

Then ask for a plumber.

Something to try, in a new chat

My kitchen drain is clogged and I'm in 94110. Use Tallyho to find a plumber who can come this week, and show me how much evidence is behind each number.

Ask in your own words. Your agent turns the request into a job spec: the system, the issue, when, a price ceiling if you have one, and the ZIP. Only the ZIP is required.

A good agent tells you how thin the evidence is. If yours quotes a likelihood without its sample size, ask for it.

find({ job })
Licensed plumbers near the job’s ZIP, ranked. Each comes with a likelihood per promise and one for the whole job, and every likelihood carries its sample size, an 80% interval, the date of its last evidence and its basis. Free, and it never contacts a business.
book({ business_id, find_id })
A handoff. Tallyho records the attempt and returns the plumber’s phone number and three questions to ask. It comes back with booked: false, because the call is yours to make.
report({ booking_id, result })
What happened, after the call to the plumber and again after the job: booked or not, the quote, when they arrived, the final price. Today, reports are what turn a starting estimate into a measurement.

03 · Status

What is real today.

  1. book

    Booking is a handoff.

    book returns the plumber’s phone number and what to ask. You, or the person your agent works for, make the call. Nothing is booked until the plumber says yes, and Tallyho doesn’t call or message any plumber yet.

  2. find

    Every number is a starting estimate.

    Until outcomes are reported for a plumber, each of their likelihoods comes back with n = 0 and basis “prior”: Tallyho’s opening guess for any plumber, not a measurement of that one. With every plumber tied, results are ordered specialists first, then nearest. On 12 November 2026, a search from 94110 found 946 licensed plumbers within 15 miles, all of them at n = 0.

  3. coverage

    Plumbing, in the Bay Area.

    Residential plumbing, for a job with a ZIP in San Francisco, Oakland, Berkeley or the Tri-Valley (Danville, San Ramon, Dublin, Pleasanton, Livermore). Any other ZIP or trade is refused, not guessed at. The plumbers come from California’s contractor license records: active plumbing licenses with a phone number, based within 15 miles of the job. Tallyho hasn’t vetted or contacted them. The license record is all it knows so far.

One likelihood from find, exactly as it comes back at n = 0.
{
  "promise": "arrives_in_window",
  "p": 0.7,
  "n": 0,
  "interval": [0.4, 0.942],
  "last_evidence_at": null,
  "basis": "prior"
}
n
The effective sample size: how much evidence is behind p. Zero until outcomes are reported.
interval
The 80% interval. It is this wide because nothing has been observed.
last_evidence_at
Freshness: when the newest evidence was recorded. null means there is none.
basis
“prior” is a starting estimate. It changes once outcomes carry the number.

The numbers get real one reported job at a time. Tell Tallyho what happened.

04 · Help

Didn’t connect?

A step here is wrong or out of date, your client isn’t listed, or a search came back odd.

hello@asktallyho.com

New to Tallyho? How it works.