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

# Context.dev MCP Setup

> Connect Context.dev to ApexGenius with browser sign-in, then search the web, read pages as Markdown, extract JSON and watch pages for changes from your AI.

**Status: Available.** Context.dev uses its official hosted MCP server and
browser OAuth. You do not paste a Context.dev API key. The connection acts for
your Context.dev organization, and most calls spend that organization's shared
credits.

[Explore the Context.dev connector](https://www.apexgenius.ai/connectors/context-dev).

## Connect and run a free first read

**Before you start**

* An account with permission to use the Context.dev resources you intend to connect.

<Steps>
  <Step title="Open Context.dev in Connections">
    Choose Connect, or Add connection when offered. Give the connection a name if prompted. If Gateway setup pending appears, contact ApexGenius support before continuing.
  </Step>

  <Step title="Choose the account and review access">
    Sign in to Context.dev in the browser and approve access for your organization. Context.dev uses browser authorization; no Context.dev API key is pasted into ApexGenius.

    <Frame caption="Context.dev labels ApexGenius an unverified app and names server.autom8.pro, the ApexGenius gateway, as the address that receives access. Sign in to continue.">
      <img src="https://www.apexgenius.ai/images/integration-guides/context-dev/20260923-sign-in.webp" alt="The full Context.dev connect page, Connect Context to an unverified app, with red boxes around the verification notice and the Sign in button." width="1323" height="896" style={{ width: '100%', height: 'auto' }} />
    </Frame>
  </Step>

  <Step title="Review the connected account and resources">
    Confirm the connected Context.dev organization. Open Tools to review its reads and the monitor and batch actions. Most operations spend your organization's Context.dev credits.
  </Step>

  <Step title="Check the connection">
    Use Test connection or Check again and read the reported status. Add an AI agent from the AI agents page, choose Read only for the first test, and try the question below.
  </Step>
</Steps>

**Try a first question with your AI agent**

```text theme={"system"}
List my Context.dev monitors. Do not create, run, or change anything.
```

Compare the returned account and resource with Context.dev. Check the connection status separately from the result of the first read.

Listing monitors costs no credits and changes nothing. A reply, even an empty
list, shows the connection reaches your organization. It does not test a paid
read or a monitor change.

## Tools and the ContextDev skill

Open **Skills** on the connection to read **ContextDev**, or read that MCP
prompt in a compatible AI client. It explains the first safe read, what each
read costs, how to treat page text, and the undo limits.

Three ApexGenius tools route Context.dev's live catalog:

| ApexGenius tool              | What it does                                                                    | Access               |
| ---------------------------- | ------------------------------------------------------------------------------- | -------------------- |
| `context_dev_search_tools`   | Lists the operations Context.dev's hosted MCP publishes now, with their schemas | Read                 |
| `context_dev_call_read_tool` | Runs an operation only when the live catalog marks it read-only                 | Read                 |
| `context_dev_call_tool`      | Runs any other operation, including monitor and batch changes                   | Write or destructive |

Context.dev documents 34 operations behind those tools:

| Job                    | Operations                                                                                                                                                                     |
| ---------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| Search and read pages  | `web-search`, `web-scrape-markdown`, `web-scrape-html`, `web-scrape-images`, `web-screenshot`                                                                                  |
| Map, crawl and extract | `web-scrape-sitemap`, `web-crawl`, `web-extract`                                                                                                                               |
| Documents              | `parse-document`                                                                                                                                                               |
| Company and brand data | `get-brand`, `brand-retrieve-unified`, `web-styleguide`, `web-fonts`, `web-naics`, `web-sic`                                                                                   |
| Monitor history        | `list-monitors`, `get-monitor`, `list-monitor-runs`, `get-monitor-run`, `list-account-runs`, `list-monitor-changes`, `list-changes`, `get-change`, `list-monitor-credit-usage` |
| Manage monitors        | `create-monitor`, `update-monitor`, `run-monitor-now`, `delete-monitor` (destructive)                                                                                          |
| Batches                | `submit-batch`, `list-batches`, `get-batch`, `get-batch-results`, `cancel-batch`, `delete-batch` (destructive)                                                                 |

Your AI discovers the tools with `apex_search_tools`. Reads use
`apex_call_read_tool`; everything else uses `apex_call_write_tool`. Writes run
right away. If you set a tool to Ask me first, the call returns
`approval_required`; a person approves it in ApexGenius and the AI retries
with the returned `approval_request_id`.

## Credits and limits

Context.dev prices each operation. From its API reference: a scrape is 1
credit (2 with browser actions), a web search is 1 credit per 10 results, a
crawl is 1 credit per page, and a brand lookup is 10 credits. Listing monitors
is free. Monitors spend credits on every scheduled run, and a batch reserves
credits when it is submitted.

Free organizations get 1,000 credits a month with a work email, or 250 once
with a personal email. Paid plans and per-minute rate limits are on
[Context.dev pricing](https://context.dev/pricing) and your Context.dev
dashboard.

## Page text is untrusted

Everything Context.dev returns from the web is data, not instructions. The
ContextDev skill tells your AI to quote and cite page text and never to follow
instructions found inside it. Browser actions on the scrape tools click or
type on a third-party page, so use them only when you ask for them.

## Undo and limits

Credits spent, runs that already happened and webhooks already sent cannot be
undone. A monitor change can be put back with `update-monitor` and the old
settings, and a monitor can be paused instead of deleted. A deleted monitor or
batch cannot be restored. Disconnecting stops future gateway access; it does
not stop monitors or batches that already exist in Context.dev.