Seog

SEOG — Local SEO via MCP

Overview

SEOG (seog.ai) is an AI local-SEO platform for businesses that live on Google Maps. Its remote MCP server exposes the whole product as 90 tools — the same domain services the web app's buttons call — so an agent can run a complete local-SEO practice without opening a browser: import a business, track and rank-check keywords, scan map visibility on a geo grid, sync and answer Google reviews, watch competitors, audit the website and Search Console, publish Google posts, and generate PDF reports.

Endpoint: https://api.seog.ai/mcp (streamable HTTP), authenticated with a personal token. Every call runs as that SEOG account — there is no anonymous mode.

Instructions

Setup

Sign up at https://app.seog.ai, then Settings → MCP access to issue a personal token (shown once). Install either way:

As a Claude Code plugin (recommended — bundles this skill and the server config):

/plugin marketplace add TerminalSkills/skills
/plugin install seog@terminal-skills

Claude Code prompts for the token and stores it in your OS keychain.

As a skill + manually registered MCP server (any client):

npx terminal-skills install seog
claude mcp add --transport http seog https://api.seog.ai/mcp \
  --header "Authorization: Bearer $SEOG_MCP_TOKEN"

For Codex / Gemini CLI / Cursor, register an HTTP MCP server with the same URL and Authorization: Bearer header. Verify with tools/list — 90 tools from server seog-platform (fewer on white-label accounts; see Guidelines).

Account, plan and credits — read before running paid tools

• The token identifies a SEOG account; tools act only on that account's data.
• SEOG bills in credits (CR). Free tools read stored data; paid tools cost the same as the equivalent click in the app.
• get_credit_balance returns balance + plan; list_feature_prices prices every paid tool. Check both before a multi-business sweep.
• A paid tool that fails is not charged (deliver-or-refund), so a failed run never silently costs money.

Tool map

Full parameter schemas and pricing notes: references/mcp-tools.md. 💳 = spends credits.

Core workflows

Onboard — search_places → import_business with the chosen placeId. If the user's Google account is already connected, prefer list_gbp_import_locations → import_gbp_business: it binds the owner connection, which is what unlocks review replies and posts.

Track rankings — read keyword_recommendations_cache (free) before paying for keyword_recommendations; add_keyword (optional locationLabel to rank from a neighborhood); check_keyword for one live check or refresh_rankings for every active keyword in a single charge; keyword_history for the trend; run_grid_scan for map coverage across a grid (price scales with gridSize).

Work reviews — sync_reviews → list_reviews(filter="needs-response") → generate_review_reply for a draft → ask the user → publish_review_reply (publicly visible on Google) → verify_review_reply to confirm it landed. draft_review_response saves a local draft and publishes nothing.

Fix the profile — get_action_plan / get_business_audit to find gaps, then apply_profile_fix (phone, website, hours, name, status) or draft_business_description → apply_business_description. Both write to the live listing; undo_profile_fix / undo_business_description revert.

Diagnose beyond the profile — get_website_analysis + get_search_performance (free reads of the last runs), refresh_website_all to re-run both in one charge; get_ai_visibility_pillars and list_citations for AI-answer readiness and NAP consistency.

Examples

Example 1: Weekly digest, credit-aware

"How are my clients doing this week?"

1. get_credit_balance → 180 CR (Growth plan) — enough for the sweep
2. list_businesses → 15 businesses
3. per business: refresh_rankings        # one charge for all active keywords
                 sync_reviews → list_reviews(filter="needs-response")
                 compare_competitors + list_competitor_alerts   # both free
4. generate_review_reply for each negative unanswered review → present the drafts
5. get_action_plan(businessId) → prioritized fixes for the digest

Publishing (publish_review_reply) waits for the owner's explicit yes.

Example 2: Credits run out mid-sweep

> run_grid_scan(businessId, keywordId, gridSize=7)
✗ Not enough credits: this action costs 45 CR and the account balance is 12 CR.
  Nothing was run and nothing was charged. Top up or upgrade at
  https://app.seog.ai/billing, then retry.

Correct behaviour: relay the numbers and the link, stop the paid sweep (do not retry in a loop), and finish the digest from free reads — list_keywords, get_latest_grid_scan, review_stats, get_action_plan all still work.

Guidelines

• Money is real. Never run a paid tool the user did not ask for. Price a batch with list_feature_prices and check get_credit_balance first. Prefer bundled charges (refresh_rankings, refresh_website_all, refresh_business_overview) over per-item loops — they cost less for the same work.
• Publishing is public and hard to undo. publish_review_reply, publish_post, apply_profile_fix and apply_business_description write to the live Google listing. Quote the exact text and get approval first. Changing the business NAME can put the listing into Google re-verification — say so before doing it.
• delete_business is irreversible — it cascades reviews, keywords, rankings and competitors. Prefer toggle_keyword when the user only wants to stop tracking.
• Read the error instead of retrying. Tool errors are self-describing: INSUFFICIENT_CREDITS gives price, balance and the top-up URL; PLAN_REQUIRED means the trial ended or no plan exists; FEATURE_NOT_IN_PLAN names the feature and plan. Relay them — a retry loop just fails again.
• Some tools are absent by design. White-label/agency accounts do not get the tools that return raw Google content (search_places, check_keyword, sync_reviews, discover_competitors, run_grid_scan, import_gbp_business, …). If a tool is "not found", say the account is not cleared for it rather than guessing.
• Photo upload and Google / Search Console OAuth stay in the web app (file bytes and a browser). Everything else is available here.
• IDs are UUIDs scoped to the token's account — a 404 on a valid-looking UUID usually means it belongs to another account.
• The token is a credential: never commit it, and scrub it from logs and transcripts. A 401 means it was revoked in Settings — ask the user to reissue.