Playbook 04

The AI SEO Tool Stack

Kimi supports MCP servers, which means it can plug directly into real SEO data: keyword metrics, Search Console, Analytics, and your Google Business Profile. This is the exact four-connection stack professionals run from the terminal.

The four connections

Plug Kimi into real data

Each connection is an MCP server. Add it to Kimi's MCP config, and the agent can pull live data instead of guessing.

DataForSEO

Market data

The workhorse. Keyword volume, difficulty, CPC, live SERP pulls, competitor gap analysis, backlink data, Lighthouse audits, and even AI-search visibility metrics, all callable by Kimi through MCP.

Use full official location names ("Australia", not "AU")
Always pull with limit + offset: 20-50 rows, never 1000
A full keyword research run should cost $0.30-$0.80. Report the cost
If the API returns null, report null. Never estimate a number and present it as data

Google Search Console

Ground truth

DataForSEO tells you about the market. GSC tells you about your site. Clicks, impressions, CTR, and position per query and page: the source of truth for what actually ranks.

Pull 28-day windows by default, compare period vs prior period
Get the exact property string from list_properties first
If OAuth fails, stop and re-run the flow. Never retry blindly

Google Analytics 4

After the click

GSC says a page gets clicks. GA4 says whether those visitors convert or bounce. Join the two every week and your reports answer "what changed" and "what to do next".

Join GSC + GA4 into week-over-week deltas
Keep a machine-readable JSON snapshot so the next run computes deltas automatically

PostProxy

Distribution

A hosted MCP connector that gives Kimi your Google Business Profile and LinkedIn. Repurpose every blog post into a GBP post with a Learn More button, and manage review replies safely.

Always call profiles_list first. Never hardcode profile IDs
Verify every publish with post_status afterwards
Auto-publish positive and neutral review replies. Queue negative ones for human approval, always

Config template

The MCP config, ready to adapt

The mcpServers JSON shape is the same across agentic tools. Swap in your own credentials and paths. Never commit real keys to a public repo.

MCP Server Config (template)

{
  "mcpServers": {
    "dfs-mcp": {
      "command": "npx",
      "args": ["-y", "dataforseo-mcp-server"],
      "env": {
        "DATAFORSEO_USERNAME": "your-dataforseo-login",
        "DATAFORSEO_PASSWORD": "your-dataforseo-password"
      }
    },
    "gsc-mcp": {
      "command": "/path/to/mcp-gsc/venv/bin/python",
      "args": ["/path/to/mcp-gsc/gsc_server.py"],
      "env": {
        "GSC_OAUTH_CLIENT_SECRETS_FILE": "/path/to/client_secret.json",
        "GSC_SKIP_OAUTH": "false"
      }
    },
    "analytics-mcp": {
      "command": "/path/to/analytics-mcp",
      "env": {
        "GA4_PROPERTY_ID": "your-ga4-property-id",
        "GA4_CLIENT_SECRET_PATH": "/path/to/client_secret.json"
      }
    }
  }
}

// PostProxy is a hosted connector: get your MCP endpoint + API key
// from the PostProxy dashboard and add it as a remote MCP server.
// Verify with auth_status and profiles_list after connecting.

Why it works: This is a sanitized template. Keep your real credentials in local config files only, and never paste them into shared chats or repos.

Hard rules

Seven operating principles that override everything

Each rule here exists because someone broke it once and paid for it.

Real data only

Never invent a metric, ranking, volume, or audit score. Three real numbers beat twenty plausible ones.

Plan before executing

Any task with 3+ steps gets a written plan first. If it goes sideways, stop and re-plan.

Verify before claiming done

"I made the change" is not "the change works". Fetch the live page, check the 301, validate the schema.

One task per session

Delegate heavy research to subagents with their own context. Keep the main context clean.

Be economical with paid APIs

Cache what you pull, dedupe before you fetch, report cost when a run is nontrivial.

No em dashes in content

Colons, commas, or separate sentences. In titles, meta descriptions, body copy: everything.

Simplicity first

Touch the minimum surface. Never speculatively rewrite a page that is ranking.

Put it to work

The weekly report, one prompt

With the stack connected, this is the prompt that runs your Monday morning.

Weekly GSC + GA4 Delta Report

Using the GSC and GA4 MCP connections:

1. Pull the last 28 days of GSC search analytics for [property] and the prior 28 days.
2. Pull the same two windows from GA4: organic sessions, engagement rate, conversions by landing page.
3. Join them and report week-over-week deltas: clicks, impressions, CTR, average position, top 5 movers up and down, conversions from organic.
4. Flag every (page, query) pair sitting in positions 11-20 with high impressions: these are striking-distance wins.
5. End with a 5-item workplan for next week, ordered by expected impact.
6. Save a machine-readable JSON snapshot of this run so next week's report can compute deltas automatically.

Why it works: The report answers 'what changed'. The workplan turns it into 'what to do next'. The JSON snapshot means no run ever re-pulls old data.

Cost discipline

Typical DataForSEO costs: a SERP pull around $0.002, Lighthouse around $0.05 per URL, content parsing $0.001-0.01 per page. Cache everything, dedupe before you fetch, and enrich only ambiguous cases. The failure mode of agentic SEO is drowning in API responses: pull less data, think harder.

End every session by writing state to disk

Keyword bank, content queue, reports. The next session, or the next agent, picks up without redoing anything. That is how a one-person operation compounds like an agency.

Want this stack set up for you, step by step?

The AI Ranking community teaches the full AI SEO workflow live: tool connections, content systems, and weekly routines, with coaching when you get stuck.

Learn how to do AI SEO Join the free community

AI Ranking on Skool: 7,400+ members learning AI search optimization. Free to start.