Documentation

The dashboard, REST API, and MCP operate the same authenticated workspace.

Install the agent extension

This configures the remote MCP server for Claude Code, OpenCode, and Codex CLI. It also removes capture hooks left by old Hive Rank versions.

$ npx hive-rank

Authenticate a workspace

  1. Run npx hive-rank auth.
  2. Open the displayed URL, enter the one-time code, sign in, and approve the workspace.
  3. Restart the client and ask the agent to run hive_whoami.

The device-login flow configures supported clients without copying a token. For CI or another non-interactive terminal, create a manual token in API + MCP settings and run npx hive-rank auth --token <token>.

Manual MCP URL

https://mcp.hive-rank.com/mcp

For Cursor or another compatible client not supported by the installer, create a workspace-wide or project-scoped token in API + MCP settings and send it as Authorization: Bearer hr_live_…. Tools and resources are discoverable at runtime.

One product, three controls

DashboardHuman review, configuration, approvals, billing, and access.
REST APIVersioned operations for integrations. OpenAPI 3.1 is available at https://mcp.hive-rank.com/api/v1/openapi.json.
MCPThe same project, prompt, collection, evidence, reporting, and action operations for agents.

Measurement methodology

Hive Rank keeps observations and inferences separate so a result can be inspected and reproduced.

Unit
A successfully collected answer run for a saved prompt and configured provider.
Evidence retained
Prompt, complete answer, provider, model and collection metadata, citations, competitors, and supporting run identifiers.
Comparisons
Visibility and competitor rates use observed answer runs. Low samples are labelled. Change is reported against a defined baseline window.
Limits
AI answers and sampled search positions vary. Reports describe observed associations and do not claim that a content change caused an outcome.

Our Lindy study mapped 2,421 discoverable URLs and inferred 2,406 unique page targets. It separated editorial and programmatic content before sampling search positions.

The study is non-customer research from a fixed repository snapshot. It also stored live Hive Rank observations.

Ask in natural language

“Create a project for Acme and add our launch prompts.”

hive_project_create · hive_prompt_create

“Preview our comparison prompts across connected providers.”

hive_run_preview

“Show where Acme appears, who appears instead, and which sources support the answers.”

hive_visibility_report · hive_citation_report · hive_competitors_report

Safety and privacy

  • Each operation requires a sign-in and stays within its workspace.
  • Hive Rank shows provider readiness and estimated cost before a run.
  • Publishing, permission changes, and actions that remove data require confirmation.
  • The local installer does not capture unrelated searches, conversations, or browser activity.