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

# Connect your agent and run a first task

> Connect to the AgentLinkOps private pilot with browser approval, inspect your projects, and choose a first campaign task or a repository-local link check.

Start with your existing agent and a page you want to earn links to. Hosted access requires a pilot invitation. A supplied local toolkit or development checkout can keep a ledger and run local checks without an AgentLinkOps account.

## Connect to the hosted pilot

Give your agent this prompt:

```text theme={null}
Set up AgentLinkOps using https://agentlinkops.com/SKILL.md.
Ask me to approve the connection, then list my projects.
```

The HTTP MCP endpoint is:

```text theme={null}
https://app.agentlinkops.com/mcp
```

For Claude Code, register the endpoint:

```bash theme={null}
claude mcp add --transport http agentlinkops https://app.agentlinkops.com/mcp
```

Complete sign-in and consent in the browser when your client requests it. The grant controls workspace access and tool scopes. Keep tokens in the client's credential store or the documented local environment; do not paste them into the conversation. MCP OAuth grants and REST credentials have different purposes.

After connecting, ask the agent to call `list_projects`. Confirm the intended project before creating a watch or importing data. If access fails, check the pilot invitation, selected account, consent and requested workspace before retrying. Client support varies; a listed endpoint alone does not establish a successful connection.

## Choose a first task

```text theme={null}
Use this project and my backlink export to prepare a shortlist for [target URL].
Keep source provenance, check selected candidates, and show why each publisher
might want to cite the page. Use your browser for questions the saved evidence
cannot answer. Save the draft plan locally for my review.
```

For a supplied placement, start with `monitor_link`, request a check with `request_link_check`, then read the job and history. Use the returned identifiers. `get_public_contacts` reads saved HTML for a watch, so check the page before requesting contacts.

See [campaign workflows](/campaign-workflows), [evidence states](/evidence) and the [tool reference](/mcp-tools/generated/overview) for the inputs and limits.

## Run the local development CLI

<Note>
  These commands require an authorized Linktrail development checkout with its dependencies installed, or the corresponding supplied toolkit. They are not a public npm install command. Public package and marketplace distribution remain separate from local validation.
</Note>

The development checkout requires Node.js 22 or later. From that checkout, inspect the CLI:

```bash theme={null}
npm run linktrail -- --help
```

To keep the ledger in your own project, run the CLI by its absolute path from that project's directory. Replace the checkout path and example URLs:

```bash theme={null}
node /path/to/linktrail/cli/linktrail.mjs init
node /path/to/linktrail/cli/linktrail.mjs add   --source=https://publisher.example/guide   --target=https://your-site.example/guide   --scope=exact --intent=wanted
node /path/to/linktrail/cli/linktrail.mjs check
node /path/to/linktrail/cli/linktrail.mjs status
```

`check` fetches public pages from your machine using the verifier shared with the cloud. It requires network access, but no AgentLinkOps account. It does not install a scheduler. Read [the ledger guide](/ledger) before committing or syncing the files.

Use `doctor` to inspect the ledger, verifier, cloud reachability and configured token. A local-only setup can still report missing cloud access. Read the [capability status](/capability-status) before treating a local command as a hosted feature.

Return to the [AgentLinkOps overview](/introduction) for the complete campaign flow.