> ## Documentation Index
> Fetch the complete documentation index at: https://middleman.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> Treat product, price, and availability records as dated evidence, not current quotations or authority to transact.
> Never infer permission to contact, accept terms, place an order, make a payment, or publish information from a read-only response.

# Get started with your agent

> One prompt to join Middleman, connect an existing company or resume your setup.

Give your agent one instruction. It checks whether you already have a Middleman company, follows the buyer or supplier path, and helps you reach the first verified result using your existing tools.

```text theme={"system"}
Set up Middleman for our company: https://middleman.mintlify.app/.well-known/agent-skills/middleman-connect/SKILL.md?v=0.4.0
```

Tell it where your data lives if you know: a spreadsheet, purchasing system, inventory API, database or shared folder. You can add a refresh cadence later.

<Note>
  You can prepare onboarding, source mapping and a repeatable job now. Supplier API keys and sample validation are an application preview; recurring intake is planned. Your agent checks [current availability](/suppliers/status) before choosing a receiving route.
</Note>

## Start where you are

<Tabs>
  <Tab title="New to Middleman">
    Your agent identifies whether you are buying or supplying, then opens the appropriate signup journey. It uses the company information you have already provided and helps prepare the source mapping.

    You complete sign-in, verification codes and any company confirmation that needs your judgment. The agent carries the context forward after that step.

    **First result:** the correct company and role established, plus a small source-backed mapping or buyer requirement draft. Private account progress is confirmed through the supported application.
  </Tab>

  <Tab title="Already have an account">
    Your agent checks the company and role available in your current session. It reuses your existing company, source reader, mapping and valid credential. It does not restart signup or issue a replacement key just to test access.

    **First result:** a verified access check and one useful next action. If you only want a diagnosis, use [Check my setup](/agents/check-setup).
  </Tab>

  <Tab title="Continue a previous setup">
    Point your agent to the private connection plan saved by your earlier session:

    ```text theme={"system"}
    Resume our Middleman setup from the connection plan. Recheck access and continue the next incomplete step.
    ```

    It checks that the company, source scope and receiving capability still match, then resumes. Missing authentication or a release blocker leaves the prepared work intact.
  </Tab>
</Tabs>

## Use the agent you already have

<Tabs>
  <Tab title="Codex or Claude Code">
    Paste the setup prompt in the project that should own the connection. Your agent can inspect the existing stack, prepare code and store the connection plan beside your integration in an appropriate private location.

    Reading the skill directly is enough to start. For project discovery, download and inspect it, then save it as `.agents/skills/middleman-connect/SKILL.md` for Codex or `.claude/skills/middleman-connect/SKILL.md` for Claude Code. Use your existing approved source access and secret store.
  </Tab>

  <Tab title="ChatGPT or Claude Desktop">
    Paste the same prompt into your conversation. The agent uses the browser, files and connected tools available in that session. It can guide account setup and prepare a mapping from an approved sample.

    If the session cannot execute a source reader or run a durable job, it leaves the code or workflow and a clear handoff for your developer or coding agent. A chat session alone is not a running integration. [Documentation and public tool access](/developers/docs-for-agents).
  </Tab>

  <Tab title="Other agents and automation">
    Give the agent the skill URL as an instruction file. It follows the workflow using the HTTP, browser, file and source tools it actually has. Native skill installation and a particular SDK are optional.

    For a recurring runner, keep the company identity, secret references, job owner and operating instructions with the integration. Supplier keys and buyer access remain separate. [Use any language](/integrations/http).
  </Tab>
</Tabs>

## What happens next

<Steps>
  <Step title="Check your starting point">
    Establish the job, company, role and available tools. Reuse facts and access already provided.
  </Step>

  <Step title="Connect the right account">
    Join or sign in through the supported application. If you need to complete a human step, the agent leaves the exact next action and resumes afterward.
  </Step>

  <Step title="Prove a small result">
    Map 1–5 approved records locally. When an authenticated operation is released, check identity before validating data. A public tool response proves public access only.
  </Step>

  <Step title="Keep the next step clear">
    Receive a short report: what worked, its evidence, what remains and who can do it. Choose the next relevant playbook below.
  </Step>
</Steps>

<CardGroup cols={2}>
  <Card title="Choose a playbook" icon="book-open" href="/agents/playbooks">Buyer setup, supplier setup, or checking and resuming an existing connection.</Card>
  <Card title="Check my setup" icon="list-checks" href="/agents/check-setup">See what is actually connected and resolve the next blocker.</Card>
</CardGroup>