⚡ How it works
- Your AI researches. The Claude app at claude.ai, or a coding agent
(Claude Code, Codex, Cursor), connects to DispatchSEO over MCP,
reads the served playbook, and mines keywords from your Search Console
data, Google Autocomplete, and what it already knows about your product.
Ideas land in a queue with the reasoning attached.
- You approve, or don't. Each idea is a card on the dashboard: the
keyword, why it's winnable, the angle. Approve, reject, or reorder.
Prefer full autopilot? Flip on auto mode and skip the queue entirely.
- The pipeline builds. Every morning a scheduled job picks the oldest
approved idea and builds it into a real pull request against your site's
repo: a guide or a small free tool, checked against the live SERP and run
through a sameness reviewer so page twelve doesn't read like page three.
- It tracks what happened. Daily rank checks, hourly Search Console
snapshots, index verification, and a journey view that tells you which
SEO stage you're actually in. When a scheduled job breaks, you get a red
banner and an email instead of silence.
The backend is deliberately boring: state, scheduling, and an approval gate.
The thinking happens in your agent, where your product knowledge already
lives.
🤖 Which coding agent
DispatchSEO talks to your agent over MCP, so "which agent" is a real question
with a real answer — including the parts that aren't finished yet.
| Agent | Connect + every tool | Unattended overnight builder |
|---|
| Claude Code | ✅ | ✅ |
| Codex | ✅ | ✅ |
| Cursor | ✅ one-paste connect, all 61 tools | ✅ via its API key (any plan mints one) |
| Gemini CLI · Copilot · any MCP client | ✅ | ❌ |
Connecting is one paste and gets you the whole tool set: research, the queue,
approvals, backlinks, reports, and building a guide when you ask for one. The
overnight builder is the separate thing — scheduled jobs that run an agent with
nobody watching — and all three do that. Every workflow carries every agent
and asks the dashboard which to run, so switching agent takes effect on the
next scheduled run with no repo change.
The honest difference is who pays: Claude Code runs on a subscription you
already have, Cursor runs on your Cursor plan's API key, and Codex is
metered by OpenAI per run.
You can also skip coding agents entirely. Add DispatchSEO to the ordinary
Claude app at claude.ai as a custom connector and your Claude does the research
and the writing on the subscription you already have, then hands the article
back for the backend to check, finish and publish. It won't build interactive
tools or run the overnight builder, and claude.ai has to be able to reach your
install, so that path wants a public URL rather than a laptop. ChatGPT can't
connect yet.
docs/AGENTS.md has the details, the support tiers, and how to
add an agent.
✅ What you need before you start
- Somewhere to publish. Either WordPress you host yourself, connected once
from Settings with an application password and posted to directly, or your
site's source in a GitHub repo, where content arrives as pull requests. Wix,
Squarespace, Shopify, Webflow and Ghost have neither door and won't work.
- An AI to drive it. Your AI is the engine, and it can be the ordinary
Claude app at claude.ai or a coding agent. Claude Code runs on the Claude
subscription you already pay for; Codex and Cursor do everything Claude Code
does here, including the overnight builder. Codex is metered by OpenAI per
run, and Cursor builds draw on your Cursor plan's included usage. See the table above.
- A machine with Docker (~1 GB RAM). A laptop works for a test drive, but
a machine that stays on is much better for daily use - a $5 VPS, a Raspberry
Pi, a desktop that never sleeps. Schedules only run while the machine is awake.
- Google Search Console access to your site.
That's the whole list. No API keys to buy, and no account on our side.
📖 Documentation
Everything is at dispatchseo.com/docs - every
page, every setting, every tool.
Pointing an agent at this? Add .md to any docs URL for clean markdown, or
fetch the entire documentation set in one request:
llms-full.txt. There's an index at
llms.txt and a SKILL.md that
walks an agent through connecting a site.
Stuck during setup? The
troubleshooting page covers the
errors people actually hit. If yours isn't there, ask in the
Discord or in
Discussions - questions get
answered and usually turn into a docs fix.
🗓️ Using it day to day
Once the wizard is done, the loop is small on purpose:
| When | You do | It does |
|---|
| Once a week, 5 minutes | Open the dashboard, work through the queue - approve, reject, reorder | Refills the queue with researched ideas and the reasoning behind each one |
| Every morning | Nothing | Builds the oldest approved idea into a pull request on your repo |
| When a PR lands | Review and merge (or let auto-merge do it) | Logs the page, requests indexing, starts tracking its keyword |
| Whenever you want | Ask your agent: "research keywords for me", "what should I write next?", "how are we ranking?" | Answers from live data over MCP - same state the dashboard shows |
The dashboard's Home page always names the next action, so you don't have to work
out what to do when you open it.
🧰 What's in the box
- MCP server with the full tool set: the
queue, keywords, rankings, pages, GSC stats, backlink prospects, trend
topics, site profile. Anything the dashboard can do, your agent can do over
MCP; parity between the two is a hard rule in this codebase.
- Trend radar: scan for rising topics in your niche, expand a topic into
concrete guide angles, and queue the good ones.
- Guide and tool builders: guides publish at most one per day, flat and
permanent (so a queue of thirty approved ideas doesn't become thirty posts
in a week); free-tool ideas build on approval.
- Backlink playbook: a prospect list prefilled with your product's copy,
tracked per submission.
- Multi-site: one deployment manages any number of sites. Each project
gets its own MCP token, its own data, its own settings.
- A password-gated dashboard for the one human in the loop.
💵 What it costs to run
Nothing, unless you want paid data. The tiers stack:
| Tier | Price | What you get |
|---|
| Search Console only | $0 | Rankings from GSC, keyword ideas from Autocomplete plus your own impression data |
| + SerpApi free key | $0 | Live SERP checks, real positions weekly (250 free searches/month) |
| + DataForSEO | pay per call | Search volume, keyword difficulty, domain rating |
Free mode finds keywords you can win. Paid mode also knows which ones are
worth winning.
☁️ Cloud version
There's a hosted version for people who'd rather not run a machine: we host it,
bundle the SERP + volume data into one bill, and replace the Google
service-account ritual with one click. It's at
dispatchseo.com. Self-hosting stays feature-complete
either way - the cloud sells convenience, not capability.
🧑💻 Developing from source
git clone https://github.com/NeoZi12/dispatchseo
cd dispatchseo
pnpm install
cp .env.local.example .env.local
pnpm dev
This is the contributor path, not the way to self-host - that's the one-command
Docker install behind the button at the top. Fill in .env.local (Supabase +
the three secrets) before starting, then open the dashboard on
localhost:3000.
pnpm build is the typecheck - run it before opening a PR. There is no
separate lint or test setup.
🏗️ Architecture, briefly
Next.js App Router, Postgres for state (a bundled container when self-hosted,
Supabase in the cloud version), mcp-handler for the MCP server at /api/mcp.
Schedules and builds run in-stack (cron + builder containers) or on GitHub
Actions. One deployment is multi-tenant: the MCP bearer token selects the
project, crons loop over all projects, the dashboard switches with a cookie.
Architecture covers this properly.
CLAUDE.md has the full conventions; it's written for agents, which
turns out to make it decent documentation for people.
🤝 Contributing
Issues before PRs, and you must understand every line you submit, including the
AI-assisted ones. Details in CONTRIBUTING.md. Questions go to
Discussions or the
Discord; vulnerabilities go through
private reporting.
📄 License
AGPL-3.0. Use it, self-host it, fork it. If you run a modified
version as a service, share the source. That's the whole deal.
