> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.eyelevel.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.eyelevel.ai/_mcp/server.

# GroundX Agent Harness

GroundX Agent Harness is a ready-made pack of GroundX know-how you install into
your AI agent (Claude, Codex, Cursor, and others). Once it's installed, your
agent already knows how to load documents, search them, pull out structured
data, and work with GroundX the way GroundX recommends, instead of guessing.
Install it, connect a GroundX API key, and you direct the agent in plain
language while it does the GroundX work.

Not sure yet whether Agent Harness is right for you? Start at [Choose Your GroundX Path](/documentation/get-started/choose-your-ground-x-path). Ready to
install? See [Install GroundX Agent Harness](/documentation/agent-harness/install-ground-x-agent-harness).

## What It Is

Installing Agent Harness into a supported agent (Claude Code, Claude Desktop,
Codex, and others) gives the agent:

* step-by-step GroundX instructions for how to do each kind of task
* reference material for GroundX's concepts and features
* the judgment to pick the right approach for the job you ask about, whether
  that's loading and searching documents, extracting data, planning a
  self-hosted deployment, or answering a question about how GroundX works
* checks so the agent verifies its own work instead of guessing

The upshot: your agent doesn't have to relearn GroundX from scratch every
session. GroundX's own guidance is built in.

## What It Can Help With

* **Loading and searching documents.** Adding documents, checking when they're
  ready, searching them, and organizing them: the pieces behind answering
  questions from your own documents.
* **Connecting the tools.** Getting your agent connected to GroundX's hosted
  tools and using them.
* **Pulling out structured data.** Deciding which fields you want, running the
  extraction on your documents, and checking the results against what you
  expected.
* **Sorting documents.** Grouping documents by type or purpose so the right
  thing happens to each one next.
* **Reports.** Asking a set of questions across your documents and assembling
  the answers, each one backed by its source, into a report you can review or
  build on.
* **Self-hosted setup.** Planning a GroundX deployment you run yourself,
  including sizing and configuration for private or air-gapped environments.
* **How GroundX works.** Explaining the product and how it fits your use case.
* **Using the SDKs.** Working with the GroundX SDKs from your own code.

## What You Install, And How The Agent Reaches GroundX

There are two parts: the know-how you install, and the way the agent actually
reaches GroundX.

* **The skills.** This is the GroundX know-how: instructions, references, and
  workflows. Installing the skills is what "Agent Harness" is. Install them
  first. (For Claude and Codex the skills come as a one-command plugin; other
  agents install the same skills however they support Agent Skills.)
* **A way to reach GroundX.** The skills tell the agent how to work with
  GroundX; to actually do the work it has to reach the GroundX API. There are
  two ways, and you can use either:
  * **GroundX's hosted tools (MCP).** Connect them once and the agent calls them
    directly. You sign in through your agent (a browser sign-in, or your API key
    in the connection settings), so you never paste a key into chat. Use this
    when your agent supports connecting remote tools.
  * **The SDK or REST API.** The agent uses your GroundX API key
    (`GROUNDX_API_KEY`, set wherever your agent runs) to call GroundX from code.
    This works with any agent that can run code or make web requests.

You don't have to set both up. Connect the hosted tools if your agent supports
them; otherwise the agent uses your key with the SDK.