/altorank

name: altorank
description: "Run AltoRank through its hosted MCP server: readiness audits, Search Console, keyword research, and SEO drafts into a review queue."
category: seo
risk: safe
source: "https://github.com/AltoRank/altorank/tree/main/plugins/altorank/skills/altorank"
source_repo: AltoRank/altorank
source_type: official
date_added: "2026-10-16"
author: AltoRank
tags: [seo, geo, mcp, content, search-console, keyword-research]
tools: [claude-code, codex, cursor, gemini]
license: "AGPL-3.0"
license_source: "https://github.com/AltoRank/altorank/blob/main/LICENSE"

AltoRank

AltoRank (https://altorank.co) is open-source SEO content software. Its tools come from the `altorank` MCP server. You can audit whether search engines and AI assistants can read a site, read an account, create **drafts**, move planned keywords, edit a draft by find-and-replace, and pause or resume a workspace. A person decides what publishes. There is no approve, hold or delete tool and no tool that publishes a draft, and you must not look for a way around that. The one publish-side tool, `altorank_retry_publish`, only re-runs a failed publish of an article a person already approved.

When to Use

• Use when the user asks about SEO or AI-search (GEO) readiness of a site, Search Console numbers, keyword research, or drafting articles for their site with AltoRank.
• Use when the user wants missing JSON-LD proposed for a page, or a readiness report for any public domain.
• Requires an AltoRank account (hosted, or a self-hosted install) and the `altorank` MCP server connected.

Connect once

If the `altorank` tools are missing, tell the user:

• Claude Code: `claude mcp add --transport http altorank https://app.altorank.co/api/mcp`, then `/mcp` → altorank → Authenticate (OAuth in the browser)
• Claude Code plugin (bundles this skill and the server): `claude plugin marketplace add AltoRank/altorank`, then `claude plugin install altorank@altorank`
• Other MCP clients: add a Streamable HTTP server at `https://app.altorank.co/api/mcp` and sign in with OAuth. Self-hosted installs use their own origin.

Two kinds of tools

• **Any domain:** `altorank_check_readiness`, `altorank_propose_schema`,

`altorank_generate_machine_readable`, `altorank_readiness_report`. Work on any public domain, not only the person's own sites.

• **Workspace tools:** everything else. They act on the person's AltoRank

account. Write access is opt-in on the OAuth consent screen; a `forbidden` error means they did not grant it. `unauthorized` means the connection expired and the person has to reconnect.

Reading results

Every tool returns `{ ok, data | error, agent_guidance }`.

• Read `agent_guidance` first. It is written for you.
• `_human`, when present, is how to describe a record to a person. Use its

labels, not raw field names or enum values.

• Articles and keywords carry `allowed_mutations`. If a verb is

`allowed: false`, tell the person the `reason` and do not retry.

• Articles carry `editor_url`. That is the link you give a person.
• `null` means unknown or unmeasured. Say "unknown", never 0.

Preflight

1. `altorank_whoami`: confirms the connection, lists workspaces and quota. 2. One workspace: use it. Several: ask which, unless the conversation says. 3. `altorank_get_workspace`: no CMS connected means the person publishes by hand. 4. `altorank_readiness_report`: if the score is low, fixing readiness comes before writing more content.

Writing a draft

1. `altorank_list_keywords`. Prefer a keyword whose `allowed_mutations.generate_draft.allowed` is true. 2. Nothing fits: `altorank_suggest_keywords`. It spends research credits, so **ask first**. Results are candidates, not saved. 3. Agree the keyword (and optionally a title) with the person. 4. `altorank_generate_draft` with one `idempotency_key` per intended draft. Nothing is published. 5. Poll `altorank_get_article` every 30-60 seconds until `status` is `review` (about two minutes). `error` means the run failed; `generation` says why. 6. Tell the person the draft is ready at its `editor_url`, then stop.

A timeout is not a failure: the draft has usually started. Repeat the call with the **same** `idempotency_key`, never a fresh one.

Changes (need write access)

Always inspect, propose, confirm, then write.

• `altorank_replace_in_article` / `altorank_bulk_replace_in_articles`: the

first call is a preview. Show the hits, get a yes, then apply. Approved, scheduled and live articles are refused.

• `altorank_reschedule_keywords`, `altorank_remove_keywords_from_plan`: only

unwritten planned entries move. Removing keeps the keyword tracked.

• `altorank_retry_publish`: only when `allowed_mutations.retry_publish.allowed`

is true (approved, then the publish failed). Tell the person first.

• `altorank_pause_workspace` / `altorank_resume_workspace`: pause stops

drafting and publishing; resume re-plans from today.

Search Console

`altorank_gsc_performance`, `altorank_gsc_cannibalization`, `altorank_gsc_coverage` and `altorank_gsc_url_inspection` read stored data; nothing calls Google. Keep these states apart:

• `not_available`: Search Console is not connected, Google refused the stored

token (a person reconnects), or no sync has succeeded (relay the error in `agent_guidance`). No data; do not say "no traffic".

• `has_data: false`: connected, nothing synced for the window. Never report 0.
• `has_data: true` with zeros: Google measured nothing. The site is not being

shown in search.

`changePct: null` means no comparison: it is given only when both windows are wholly synced (`comparable: true`). Coverage `bucket: "unknown"` means nobody asked Google and the page was not seen in search, not "not indexed". `inspection_superseded: true` means Google served the page after an old "not indexed" verdict, so it is indexed now.

Examples

Example 1: check a site before writing

User: "Can AI assistants read example.com?"

1. Conf