rankforge

Honest, local-first SEO autopilot. Audit your site → research keywords → brief the SERP → write scored articles → weave internal links → publish → track rankings. One shared Python core, two front-ends: a CLI and an MCP server you can drive from any MCP client.

No SaaS, no account, no required API keys. Your site data and your .env never leave your machine — it's a single SQLite file you can delete and rebuild.

git clone https://github.com/Makeph/rankforge && cd rankforge
pip install -e ".[all]"          # or `pip install -e .` for the lean core
rankforge init
RANKFORGE_SITE=https://your-site.com rankforge audit
rankforge keywords research "your topic"
rankforge write "your best keyword"      # scored SEO + GEO before you ever publish
rankforge tick --dry-run                 # see what the autopilot would do

Why this exists

The "AI SEO autopilot" category (AutoSEO, Oscar AI, and friends) charges ~$150/month for a loop you can run on your own machine: pick a keyword, read the SERP, write an article, add links, publish, track. rankforge is that loop, rebuilt small and honest:

Everything degrades gracefully: no key → article generation falls back through local agent CLIs to a deterministic outline; no WordPress → markdown export; no network → audit/score/link still work on stored data. Nothing publishes without passing the quality gate, and nothing publishes at all unless you configured a target.

The engine trick

Article writing is pluggable, and three of the engines are agent CLIs you may already pay for — so generation costs nothing extra and no API key ever enters this tool:

RANKFORGE_ENGINE = auto | anthropic | codex | kimi | qwen | outline

• anthropic — the Anthropic API (ANTHROPIC_API_KEY).
• codex / kimi / qwen — driven via their non-interactive prompt modes.
• outline — offline, deterministic: expands the brief into a structured skeleton with TODOs. Never publishable, never blocking.

auto walks whatever is actually available (rankforge engines shows the order). Every draft is scored; if it lands under RANKFORGE_MIN_SCORE, the engine gets one revision pass listing the exact named checks that failed — then the article is either ready or held as a local draft.

SEO score + GEO score

rankforge score <id> prints two rubrics, check by check:

• SEO 0–100 — keyword in H1/intro/headings, structure, length vs the SERP-derived target, density, internal + external links, readability, lists.
• GEO 0–100 — will an answer engine (ChatGPT, Perplexity, AI Overviews) quote you? Question-phrased H2s, ≤60-word direct answers under them, an FAQ block, a liftable definition, lists/steps, concrete figures.

No magic, no vibes: the scorer is ~150 lines you can read, and the writer never sees it — only its plain-language feedback.

The pipeline

rankforge audit                      # crawl + on-page issues + site-wide findings
rankforge keywords research <seed>   # autocomplete → cluster → opportunity 0-100
rankforge brief <keyword>            # live SERP: target length, questions, outline
rankforge write <keyword>            # engine of your choice, scored, stored
rankforge links <id> --apply         # internal links from YOUR crawled pages
rankforge publish <id> [--dry-run]   # WordPress draft or markdown front-matter
rankforge track / report             # DuckDuckGo positions + movement over time
rankforge tick                       # all of the above, once a day, guardrailed

The autopilot's guardrails mirror the rest of the tool: RANKFORGE_DAILY_ARTICLES (default 1), RANKFORGE_MIN_SCORE (default 70, below = held as draft), WordPress posts land as drafts unless you say otherwise, and --dry-run always shows the plan first.

# typical cron line — one article per weekday morning, as WordPress drafts
0 7 * * 1-5  rankforge tick

MCP (drive it from any MCP client)

pip install "rankforge[mcp]"
rankforge mcp        # stdio server

{
  "mcpServers": {
    "rankforge": { "command": "rankforge", "args": ["mcp"] }
  }
}

Tools exposed: audit_site, research_keywords, build_brief, write_article, score_article, suggest_internal_links, check_rank.

Configuration

Everything is optional — rankforge init drops a commented env.example.

Install from source

git clone https://github.com/Makeph/rankforge
cd rankforge
pip install -e ".[dev]"
pytest          # offline test suite, no keys needed

Responsible use

rankforge is a content system, not a spam cannon: one article a day by default, a quality gate that refuses to publish weak drafts, WordPress drafts by default so a human stays in the loop, polite crawling with a real user-agent and jittered pacing, and rank checks kept to a minimum. Generated content is your responsibility — review it before it ships.

License

MIT © 2026 Aurore Biakou