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

> ## Agent Instructions
> The PartFinder API is served at https://api.partfinder.dev. Authenticate with `Authorization: Bearer $PARTFINDER_API_KEY`; create an account at https://console.partfinder.dev/sign-up and pick pay as you go or a monthly or yearly plan (see https://partfinder.dev/pricing), then create a key. Pay as you go starts with 100 free credits, once per person and per company.
> Install the PartFinder agent skill with `npx skills add https://docs.partfinder.dev` (skill file: https://docs.partfinder.dev/skill.md).
> The OpenAPI document at https://api.partfinder.dev/v1/openapi.json is the source of truth for request and response schemas.
> What each call costs and current plans: https://api.partfinder.dev/llms.txt.

# Connect your AI agent to the PartFinder MCP server

> Add the PartFinder MCP server to your AI app and sign in: the server address, the two permissions you approve, what each tool call costs, how jobs are read, how to disconnect an app and how to use your own API key instead.

The PartFinder MCP server gives an AI agent the PartFinder API as tools. Your agent then
calls PartFinder itself when a question needs it: stock and lead times for a part, its supply risk,
alternates, a BOM scan, disruptions by country or a request for quote.

## Connect and sign in

The server address is:

```text theme={null}
https://api.partfinder.dev/mcp
```

1. Add the server address in your AI app.
2. The app opens the PartFinder sign-in page. Sign in with your
   [console](https://console.partfinder.dev/sign-up) account.
3. Choose the organization the app works for.
4. Approve the two permissions below.

Any AI app that supports signing in to an MCP server works. You need no API key. If you are not in an
organization yet, create one in the console first, or ask an admin to add you to one.

These pages show the steps for one app each:

> [card] [ChatGPT & Codex](https://docs.partfinder.dev/mcp-chatgpt-desktop.md)
>
> `~/.codex/config.toml`

> [card] [Claude Code](https://docs.partfinder.dev/mcp-claude-code.md)
>
> `claude mcp add`

> [card] [Claude and Claude Desktop](https://docs.partfinder.dev/mcp-claude-desktop.md)
>
> Custom connector

> [card] [Cursor](https://docs.partfinder.dev/mcp-cursor.md)
>
> `.cursor/mcp.json`

## The two permissions

The sign-in page asks you to approve two permissions for the organization you chose:

| Permission | What the sign-in page says |
| - | - |
| Read | Read your organization's parts research, BOMs, searches, comparisons and requests for quote, and look up parts, alternates and disruption events, which uses your organization's credits. |
| Write | Run searches, comparisons and BOM analyses, send requests for quote to suppliers and delete them, using your organization's credits. |

Read covers looking things up. A lookup such as researching a part uses credits. Write covers starting
searches, comparisons and BOM analyses, and sending or deleting requests for quote. If you did not allow
a permission, a call that needs it is refused. Connect the app again and allow it.

## Let your agent set it up

A coding agent can add the server for you. Paste this prompt into it:

```text theme={null}
Connect the PartFinder MCP server for me:
1. Add a remote MCP server named partfinder at https://api.partfinder.dev/mcp
2. Add no API key and no header. The server signs me in.
3. Then tell me the steps only I can take: approving the server, and signing in to PartFinder in my browser.
```

## What the tools are

Each tool is one API operation, named by its operation id in the
[API reference](https://docs.partfinder.dev/api-reference/overview.md): `researchPart` is `GET /v1/parts`, `searchParts` is
`POST /v1/search`, `sendRfq` is `POST /v1/rfq`.

* Tools that read are marked read-only. `deleteBom`, `deleteSearch`, `deleteComparison`, `deleteRfq` and
  `sendRfq` are marked destructive.
* `sendRfq` and `sendRfqMessage` send a real email to the supplier named in the call.
* Key and webhook operations are not tools. Set them up from your own code (see
  [Authentication](https://docs.partfinder.dev/authentication.md) and [Webhooks](https://docs.partfinder.dev/webhooks.md)).
* A file upload is not a tool. To scan a CSV or XLSX file, send it to `POST /v1/bom/upload` from your
  own code (see [Analyze a BOM](https://docs.partfinder.dev/bom-analysis.md)), or give the agent its lines for `analyzeBom`.

A tool result carries the API's JSON response. Its `summary` names the next tool to call, never an HTTP
path.

## What a tool call costs

A tool call is the matching API call. It costs the same credits, counts toward the same rate limit and
appears in the console's usage like any other call. Calls from an app you signed in with appear under
that app. `ping` is free, and a refused call is not charged. Listing the tools costs nothing. Prices per
call are at [api.partfinder.dev/llms.txt](https://api.partfinder.dev/llms.txt).

## Jobs answer with an id

`searchParts`, `compareParts` and `analyzeBom` start a job and answer at once with its `id`.

1. Call `getSearch`, `getComparison` or `getBom` with that `id` right away. Do not wait first.
2. The tool waits while the job runs. It returns when the job ends, or after 45 seconds with `status`
   `processing`.
3. While `status` is `processing`, call it again.

Reading a job costs no credits, but each read counts toward the rate limit.

## Disconnect an app

The console's **Connected apps** page lists the apps people signed in with, one row per app and person.
Organization admins see every app in the organization. Members see the apps they signed in with themselves.

**Disconnect** stops that app for that person. Every call it makes is then refused, even if it
signs in again, until someone chooses **Allow again** on the same row. A member cannot allow again an app an
admin disconnected. To stop a person entirely, an admin removes them from the organization.

## When a call is refused

A refused call comes back as a tool error. The JSON carries the API's `error` code and a `summary` that
says what to change. A call with no sign-in and no key, or an invalid one, is refused before any tool
runs, with HTTP `401`. Every code is listed in [Errors](https://docs.partfinder.dev/errors.md).

## Can I use my own API key?

Yes. An agent that cannot sign in, or a script you run yourself, can send an API key instead. Any agent
that accepts a remote MCP server with a header needs two values.

| What | Value |
| - | - |
| Server address | `https://api.partfinder.dev/mcp` |
| Header | `Authorization: Bearer <your API key>` |

Create the key in the [console](https://console.partfinder.dev/sign-up). Where your agent can read it
from your environment, keep it in `PARTFINDER_API_KEY`. A configuration file that holds the key, such as
Claude Desktop's, stays on your computer: never share or commit it. A key can call every tool. Each app
page above shows its key configuration too.

To have a coding agent set it up with a key, paste this prompt into it:

```text theme={null}
Connect the PartFinder MCP server for me:
1. Add a remote MCP server named partfinder at https://api.partfinder.dev/mcp
2. Send the header Authorization: Bearer <key>.
3. If this agent's config can read an environment variable, refer to PARTFINDER_API_KEY and never write the key into a file. If it cannot, tell me where in the config file to paste the key myself. Never ask me for the key in the chat.
4. Then tell me the steps only I can take: setting PARTFINDER_API_KEY or pasting the key, and approving the server.
```


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.