AgentLinkOps / Getting started

Connect your
backlink tools.

For your agent: read the public setup brief or browse the machine-readable index. You approve account access through OAuth.

Research backlink opportunities, compare competitor sources, verify selected candidates and monitor earned placements. Your agent keeps campaign decisions, outreach and link records in the tools you already use.

This is an invitation-only development pilot. You need an approved AgentLinkOps account to connect to the service.

Connect over MCP

Have your invited account ready. You can inspect its connections in your workspace.
In your agent's settings, add an HTTP MCP server using the URL below. Choose its OAuth connection flow.
Sign in and review the requested permissions. Approve the workspace connection, then return to your agent.
Verify the connection with get_workspace({}) and list_projects({ "limit": 20 }). Start with projects:read; these calls do not create monitors or request checks.

https://app.agentlinkops.com/mcp

Client setup varies. Use an MCP client that supports HTTP transport and OAuth. AgentLinkOps provides scoped authorization and requires explicit consent. You can also create a scoped API key in Agent access for REST integrations.

When you are ready to monitor, ask your agent to use an existing project or create one with projects:write. Provide a known source page and destination URL, then approve watches:write access to register the link. Creating a watch starts recurring checks that share your workspace allowance. Reading its record and history requires watches:read.

Scheduled checks continue when your agent is offline. Monitoring supplied URLs does not discover all backlinks on the web.

Tools with a clear job

This guide covers monitoring, discovery and competitor research tools. Discovery availability is described below. Use your connected agent’s tool list for current schemas. Permission checks apply to every call, and project restrictions apply to the records returned.

Job	Tools
Workspace & projects	`get_workspace` `list_projects` `create_project`
Placement monitoring	`list_link_watches` `get_link_watch` `monitor_link` `import_link_watches` `update_link_watch` `request_link_check` `get_check_job`
Evidence & changes	`get_link_history` `list_link_events` `get_public_contacts` `export_link_watches`
Profile & anchors	`get_link_profile` `get_anchor_report`
Destination health	`list_targets` `get_target` `monitor_target` `update_target` `request_target_check` `get_target_job` `get_target_history` `get_target_placements` `list_target_events`

Manual rechecks and scheduled source/destination checks share your workspace allowance. Reuse an idempotency key when retrying the same manual check request. Pause a monitor when you no longer want recurring checks.

Discover and verify opportunities

Backlink research starts with source pages worth investigating. AgentLinkOps stores candidates with their origin, dates and coverage, then lets your agent select a page for verification.

Pilot availability: importing candidates and requesting individual checks are available. Live discovery requests are disabled while the discovery workflow completes validation.

Where candidates come from

Your exports. Import rows from sources you already use. The local CLI reads supported supplier CSV formats; the MCP import accepts normalized rows. Imported rows stay private and do not populate the shared crawl corpus.
Our niche crawl. The crawler and reverse lookup are built and have run bounded research crawls. Connecting that corpus to live discovery requests remains in validation. Coverage is limited to pages actually observed.
An optional supplier. The design supports an authorized supplier connection. This path is deferred and requires its own setup; it is not needed to import your existing exports.

Job	Tools
Import candidate rows	`import_backlinks`
Request discovery (disabled in the pilot)	`discover_backlinks`
Read candidates and their coverage	`get_discovery_run` `list_discovery_candidates`
Check a selected candidate	`verify_discovery_candidate` `get_candidate_verification`

A candidate is a research lead. It becomes verified evidence only after a complete page check. Selecting a candidate for a one-off check does not enroll it in recurring monitoring.

Compare competitor link sources

Approve the competitors you want to compare, keep dated snapshots of their source pages, and inspect overlap by page or referring host. Gap reports retain the rows behind each result so your agent can investigate the source.

Built and tested; live workflow in validation. Reports read stored competitor inventories. New inventory requests depend on live discovery, which is currently disabled. A source missing from your selected dataset does not prove that it has no link to your site.

Job	Tools
Choose competitors	`create_competitor_set` `list_competitor_sets` `get_competitor_set` `update_competitor_set` `retire_competitor_set`
Request and save an inventory	`request_competitor_inventory` `capture_competitor_inventory`
Inspect dated source rows	`list_competitor_inventories` `list_competitor_inventory_rows`
Compare page and referring-host gaps	`get_competitor_gap_report`
Review and exclude sources	`get_competitor_exclusions` `preview_competitor_exclusions` `save_competitor_exclusions`

Use the results to choose what to investigate, then verify selected candidates. Your agent handles publisher research and any outreach you authorize. Return to the discovery tools.

Read what was observed

A placement has both a last known state and a latest check attempt. A complete observation can establish link presence, its anchor, and its attributes. A blocked or incomplete request stays unknown; it does not erase the previous verified state.

Confirmed loss requires independently repeated, complete checks. Destination monitoring has its own history: an unavailable destination does not mean the source link disappeared.

Saved HTML is private evidence. The pilot keeps raw snapshots for 30 days; observation metadata and history remain after a snapshot expires. Public contact extraction reads explicitly published emails and contact-page references from the latest saved page when that observation is complete. It does not infer addresses or establish deliverability.

Profile and anchor reports describe your monitored dataset and its evidence dates. They do not claim complete coverage of a site's backlink profile.

Keep your working context local

The project includes a Python and SQLite toolkit with Claude and Codex plugin manifests, shared skills, and campaign templates. Its CRM combines prospects, contacts, outreach references, and earned placements in your own workspace.

Your agent can use your existing browser and email tools for research and outreach, then record the outcome locally. Optional REST sync brings AgentLinkOps observations into that record.

The toolkit also creates offline HTML placement reports and manages user-selected disavow imports and exports. Costs and private notes are included in reports only when explicitly requested. Disavow decisions and Google submission remain with you.

The toolkit is currently distributed from the project repository's toolkit/plugin/agentlinkops directory. A public marketplace listing and public download release are not available yet.

Give access deliberately

Use Agent access in your workspace to inspect connections, revoke grants, and create or revoke scoped API keys. Keys are displayed once and expire within the period you select.

Keep credentials in your agent's existing secret store. An OAuth token issued for MCP is bound to the MCP endpoint; it is not a generic REST credential. The local REST connector uses a scoped API key.

Your AgentLinkOps connection grants no access to your mailbox or local files. The service does not send outreach.

Open workspace

Connect yourbacklink tools.