Know when the agents
pick your stack.

See what Claude Code picks when it searches, then query the same crowdsourced rankings from OpenCode, Codex, or any MCP client. Free to read · authenticated contributions mint credits.

Request early access → Install the MCP

# install the CLI — wires MCP into Claude Code, OpenCode, Codex
$ npx hive-rank

→ 23 skills installed
→ MCP server: mcp.hive-rank.com
→ Capture hooks enabled

# then run any hive skill inside your agent
$ claude "/hive:rankings best CRM for startups"

claude-code · /hive:rankings

user: /hive:rankings best postgres hosting

→ hive_rankings("best postgres hosting") ✓

→ hive_trending() ✓

Top domains ranked by crowdsourced observations — positions, trend direction, and contributor consensus.

The infrastructure for agent visibility.

Observe what agents search for. Attribute visibility to the frameworks riding each retrieval surface. Act on it with twenty-three agent skills.

What Claude Code picks, captured.

Every WebSearch and WebFetch Claude Code makes in a hive session is anonymized and contributed — no scraping, no schedule, you see what the agent saw. OpenCode and Codex run the same MCP tools today; their contribution hooks land next.

Claude Code hook live

OpenCode MCP only

Codex MCP only

Cursor soon

Rank + hive score.

Position alone lies. Hive Score (0–100) blends visibility, breadth, consensus, and freshness so you know whether a #1 is load-bearing or a fluke.

→ visibility — rank strength + top-spot bonus

→ breadth — across how many queries

→ consensus — contributor + observation count

→ freshness — how recent the signal is

23 skills, one MCP.

Slash commands that ship with the CLI: rankings, score, trending, competitors, audits, briefs, backlinks.

/hive:rankings — who ranks for a query

/hive:score — your domain, 0–100

/hive:competitors — who's beating you

/hive:content — optimized briefs

see all 23 →

Connect any agent
in thirty seconds.

One command wires the MCP into Claude Code, OpenCode, and Codex. Any other MCP client connects with the JSON config below.

One command

Any shell. The installer detects Claude Code, OpenCode, and Codex automatically.

# any shell $ npx hive-rank

MCP config

Any MCP-compatible client. Point it at the production server and you're live.

 { "mcpServers": { "hive-rank": { "url": "https://mcp.hive-rank.com/mcp" } } } 

Agent skills

23 slash commands drop into Claude Code's ~/.claude/commands/hive/ at install time. OpenCode and Codex access the same behavior through the MCP tools directly.

  /hive:rankings # network rankings /hive:score # 0–100 authority /hive:competitors # who outranks you /hive:content # briefs /hive:help # list all 23 

Contribute to earn

Run a hive skill, and the searches your agent makes next in that session contribute anonymized signal to the network. Authenticated contributions also mint credits.

 # the flow
/hive:<skill> → activates the session
web searches → captured + anonymized
the_hive → aggregates + normalizes
you (authed) → credits mint to ledger
privacy → SHA-256, no identity 

Your buyer stopped
clicking links.

Your next customer never visits your site. They ask an agent. The agent reads docs, picks a package, ships a PR. If your SEO tool is still counting clicks, it's counting the wrong thing.

Optimized for humans with cursors.

—
Scheduled crawlers. Ahrefs and Semrush sample the web on their timetable, not the agent's.
—
Keyword → ranking page. Counts clicks that increasingly never happen.
—
Crawler cadence. Daily or weekly rank refresh — already stale the moment an agent queries.
—
Subscription pricing. $100–$500/mo before you see your first ranking.

Optimized for agents with tool calls.

→
In-session observations. Ranking data flows from real agent searches as they happen.
→
Task → picked domain. The moment the agent chose you — or didn't.
→
Real-time network. Crowdsourced from every Claude Code operator running the contribute hook; OpenCode + Codex hooks next.
→
Free to read, earn to run more. Query the network anonymously. Authenticated contributions mint credits that pay for your own monitor probes.

The hive, right now.

Every number below is live from the production network. No vanity metrics, no projections.

Network contributors

Contributing via Claude Code's capture hook; OpenCode + Codex hooks next.

Queries tracked

Distinct searches observed across the network.

Sightings logged

Domain-level rank observations contributed.

Things we get
asked a lot.

Didn't cover something? Reach us at hello@hive-rank.com.

Is this AI SEO, GEO, or LLM visibility? +

Related, but different. Most of those tools chase answer engines. Hive Rank specifically measures how coding agents discover and pick packages, docs, and domains — currently capturing Claude Code's WebSearch/WebFetch traffic, with OpenCode + Codex hooks next. Different surface, different measurement.

Which agents does the MCP support? +

The installer auto-wires Claude Code, OpenCode, and Codex CLI. The MCP itself speaks the open protocol — Cursor, Windsurf, Continue, and any custom client can connect with the JSON config above.

What data do you actually collect? +

From the MCP contribution hook: queries, URLs, and observed positions — public data the agent already fetched. Your contributor identity is a one-way SHA-256 hash; no email, no account needed to contribute. The dashboard does use email + password (Better Auth) so you can manage tokens, monitors, and credits across devices — those two paths are separate and you can use one without the other. See the docs for the full schema.

How is this free? +

Contribute-to-earn. Once you invoke a /hive:* skill, the hook captures the web searches the agent makes next in that session and adds them to the network as anonymized signal. In return, you read rankings from everyone else's contributions. Hatchling tier hands you 500 credits on signup; authenticated contributions mint more credits back into your workspace ledger. No subscription, no card.

What's the Hive Score? +

A 0–100 authority number that blends visibility, breadth, consensus, and freshness. Position alone is noisy; Hive Score is how load-bearing that position is. Query any domain with /hive:score.

Can I opt out of contributing? +

Today the cleanest path is npx hive-rank --uninstall, which removes the hooks that contribute. To purge data you've already sent, email privacy@hive-rank.com with the first 8 chars of your contributorId (run /hive:privacy to see it). An in-product toggle lands with the read-only paid tier.

Know when the agents pick your stack.