Skip to main content

SEO Agent: an open-source SEO agent that builds your backlinks

SEO Agent

Open-source AI SEO Agent and MCP server for Claude Code, Cursor and Codex.

MIT licence GitHub stars MCP server Docs

SEO Agent is an open-source SEO agent for link building. It runs as an MCP server inside Claude Code, Cursor, Codex or any MCP client: you say what you need, it asks for your URL and keywords, plans the campaign, builds each link by following that site's method, checks that the link went live, and hands you the live URLs with proof.

What it does

  • Searches a library of 1,141 backlink sites, every one verified reachable this month with Domain Authority, referring domains, spam score and dofollow status. Twenty-five ship free inside the package; the rest show up locked, by name and DA, and unlock with a subscription (search_sites). The free plan builds ten links, one of them on a DA 90+ site.
  • Gives your assistant the method for each site: the playbook for that kind of link, the guide steps parsed into actions with the exact button names, and what the site requires (get_method).
  • Plans a campaign for your URL: sites inside your DA range, the dofollow share you asked for, methods spread, anchor text assigned by ratio (plan_campaign, subscription).
  • Builds links in your browser with a proof screenshot on login-free sites, free; on account-based sites with the model-driven executor on a subscription (build_link, queue_build).
  • Builds a Facebook album that ranks and sends you traffic, and a followed link on your own GitHub Pages site, both unattended after one sign-in. Subscription only (get_traffic, publish_page).
  • Verifies every link by fetching the live page and checking the hyperlink, the anchor text and nofollow (verify_link).
  • Handles gates in the chat: at a captcha, email code or login it asks you, or with a subscription uses a service you connected once (resolve_gate, connect_service, read_inbox).
  • Suggests the keywords and anchors by reading the target page itself, then allocates them across the five anchor types so your profile is not one phrase forty times (suggest_keywords, subscription).
  • Watches the links afterwards: weekly re-checks of everything it placed, plus every referring domain to your site, what is new and lost each month, spam flags and a disavow file (recheck_links, monitor_backlinks, subscription).
  • Finds what competitors have and you do not, with the method to build each one attached (competitor_gap, subscription).
  • Paces the campaign so links arrive steadily across days and methods rather than in a burst, and runs many sites under one subscription (register_site, subscription).

Twenty-six tools in all; the tool reference lists every one.

Quick start

pip install seoagents && playwright install chromium
claude mcp add seoagents -- seoagents

Any other MCP client, in its mcpServers config:

{ "mcpServers": { "seoagents": { "command": "seoagents" } } }

Then ask: Build a link to my site on a dofollow site from the free list, and verify it.

Documentation: https://m4mansoor.github.io/seo-agent/ · Repository: https://github.com/m4mansoor/seo-agent

A session, start to finish

A session: the user asks for 5 dofollow backlinks at DA 60 or higher, the agent asks for URL and keywords, plans from the free list, builds on Medium, IMDB, Zotero, ArtFire and ouo.press, and verifies each

That is the whole interaction, on the free list alone. In detail:

  1. You ask in plain language: I need 5 dofollow backlinks, DA 60 or higher.
  2. The agent asks for two things: the URL the links should point to, and your main keywords. It stores your sign-up identity once so every form is filled the same way.
  3. It plans: picks sites by authority and effort inside your DA range, keeps the dofollow share you asked for, spreads the methods so it is not five profile pages, and assigns anchor text by ratio: exact, partial, branded, naked, generic. The free list reaches DA 96: Medium, IMDB, Zotero, ArtFire, Milkyway (.edu), all dofollow. When you ask for more at the top end, it shows what the full library holds and offers the subscription.
  4. It builds: login-free sites are built by scripted playbooks with a proof screenshot. Sites that need an account, which is where the high DA lives, are built by the model-driven browser executor: it reads the site's playbook, the guide steps with the exact button names, and a snapshot of the page each turn, then signs up, fills the profile or writes the post, and places the link. In free mode it hands you the same method to follow in your own browser. Long builds run in the background so your chat never times out.
  5. It verifies: opens the public page and checks that the target is a real hyperlink, that the anchor matches, and whether the link is nofollow. Plain-text URLs, redirect pages and noindex pages do not count as verified.
  6. It reports: live URL, anchor, dofollow or nofollow and proof per link, a campaign report, and a re-check later so you know which links still stand. When a site asks for a captcha, a verification code or a login, the agent asks you in the same chat and continues once it is cleared.

The session above is illustrative.

How it works

Plan, build, verify, report

What is in the library

Sites 1,141, verified reachable this month. The catalogue holds 1,248; 107 failed the last check and are hidden from search and planning until they come back
Dofollow 900
DA 90 or higher 117
Metrics per site Domain Authority, referring domains, total backlinks, spam score, organic traffic
Methods profile fields, articles and guest posts, social bookmarking, web 2.0 pages, forums and signatures, directories, shared documents, comments, Q&A, URL shorteners
Per site the step-by-step method parsed into actions, the placement step, and what the site requires: account, inbox, captcha, social login, upload, moderation

Twenty-five of these sites, with their methods, are bundled free, and the free plan builds ten links on them, one of which may be on a DA 90+ site. The rest of the library, the campaign planner, automatic building and verification, Facebook albums and GitHub Pages links come with a subscription.

Free: 25 sites from GitHub, right now

The package is the agent. It runs on your machine, builds links in your own browser on the login-free sites, follows the method with you on the rest, verifies every link and keeps the log. No account, no key, nothing sent anywhere. Requires Python 3.10 or newer.

Claude Code

pip install seoagents && playwright install chromium
claude mcp add seoagents -- seoagents

Claude Desktop, Cursor, Codex and any other MCP client take the same command in their config:

{ "mcpServers": { "seoagents": { "command": "seoagents" } } }

Then ask: I need 5 dofollow backlinks from the free list for https://example.com, keyword "example". The assistant searches the 25 sites, calls build_link where it can build by itself, follows get_method where an account is needed, verifies each link and logs it. Ask how many links have I built any time; it calls account.

Where the work happens

Two kinds of tool, and the difference matters.

On your machine, always. Anything that drives a browser: building a link, building a Facebook album, checking a link is live. It uses your own browser and your own signed-in accounts. Your passwords are never sent anywhere, no session of yours is ever held on a server, and it keeps working even when our service does not.

From the service. The library of 1,248 sites, the campaign planner, keyword suggestions, the backlink monitor, competitor gap and reports. That is where the data lives.

Subscribe: every site, $97 a year or $27 a month

When you ask for more high-DA sites than the free list holds, the assistant shows what is locked and gives you the payment links; or call upgrade. Subscribe at https://mcp.seoagents.dev/buy?plan=yearly ($97 a year, about $8 a month) or https://mcp.seoagents.dev/buy?plan=monthly ($27 a month, cancel any time), copy the key from the success page, and tell your assistant:

activate le_your_key

No restart, no config file. From that message on, the same MCP has the full 1,141-site library, the campaign planner, browser building on account-based sites, gates with connected services, background jobs, identity generation, monitoring, campaign reports and a dashboard. Fair use is 500 placed links per key in any 30 days.

No pip? Use the link instead. The success page also shows a personal MCP link, https://mcp.seoagents.dev/u/le_…/mcp. Paste it into Claude Desktop or claude.ai (Settings, Connectors, Add custom connector, no OAuth), ChatGPT (Settings, Connectors, Developer mode) or Cursor ("url" instead of "command"). It carries your key, so keep it private.

What it replaces

Fifty verified, dofollow-checked backlinks a month is a job. Typical ways to get it done, at typical market rates:

Cost Time to 50 links Knows the sites?
In-house link builder $2,500 to $4,000 a month salary, plus tools 3 to 6 weeks, at 30 to 60 minutes per link Only the sites they have used before; a new hire starts from a blank list
Agency or marketplace $150 to $400 per DA 40+ dofollow link, so $7,500 to $20,000 per 50 4 to 8 weeks Their list, not yours; you never see the method
Freelancer on a gig site $10 to $50 per link 1 to 2 weeks Usually the same 100 sites everyone else spams
SEO Agent $97 a year or $27 a month An afternoon; login-free sites build in seconds each 1,141 sites with the exact method for each, re-verified monthly, with DataForSEO metrics

The library is the part a team cannot copy quickly. Every site in it came from years of link building: which DA 60 profile page still gives a dofollow link, which forum strips links from new members, which shortener sits behind a captcha, which directory approves in a day. Each entry carries the step-by-step method with the exact button names, what the site requires, and referring domains, spam score and traffic from DataForSEO. A person could rebuild that list, but it would take them the same years.

The agent does not replace judgement: you still choose the keywords, the anchor ratio and which links are worth having. It replaces the hours, and it remembers the sites.

Twenty-five sites from the library, all verified reachable, DA 20 to 96, 25 of them dofollow, two of them at DA 90 or higher. Each has a method your assistant can follow. The free plan builds ten links on this list, one of which may go on a DA 90+ site.

Show all 25
Site DA Referring domains Link Method Needs
Medium 96 1,371,100 dofollow Article or blog post sign-up
IMDB 95 891,904 dofollow Profile website field no account
Zotero 75 51,364 dofollow Bookmark / link submit sign-up
ArtFire 75 31,486 dofollow Profile website field sign-up
Milkyway (.edu) 75 9,011 dofollow Profile website field sign-up
DsiBlogger 74 37,269 dofollow Article or blog post sign-up
FireBlogz 74 39,401 dofollow Article or blog post sign-up
Beatstars 73 39,465 dofollow Page builder site sign-up
Flip HTML5 73 226,903 dofollow Shared document sign-up
Figma 71 101,339 dofollow Social post sign-up
MD Anderson 71 918 dofollow Other no account
Wakelet 71 191,490 dofollow Bookmark / link submit sign-up
GrowthHackers 68 7,690 dofollow Forum post sign-up
Gust 67 11,264 dofollow Directory listing sign-up
RiseUp 64 11,528 dofollow Page builder site sign-up
Wantedly 64 29,585 dofollow Bookmark / link submit sign-up
XtGem 64 26,727 dofollow Page builder site sign-up
Ouo Press 60 4,724 dofollow URL shortener no account
BlogFree 58 5,019 dofollow Forum signature sign-up
Pub HTML5 56 99,088 dofollow Shared document sign-up
AxMag 54 1,668 dofollow Shared document sign-up
Yooco 43 8,568 dofollow Comment sign-up
N9.cl Shortener 40 17,321 dofollow URL shortener no account
Goolnk.com Shortener 23 1,922 dofollow URL shortener no account
YellKey Shortener 20 389 dofollow URL shortener no account

More lists, with DA and referring domains: social bookmarking sites, profile creation sites, guest posting sites, web 2.0 sites, directory submission sites, forum posting sites, high DA backlinks.

Why SEO Agent and not another SEO MCP server

  • It builds, not only reports. Most SEO MCP servers wrap a metrics API. This one follows a site's method in a browser, places the link and proves it with a screenshot and a live-page check.
  • The library is the moat. 1,141 sites, each with its parsed step-by-step method, re-verified reachable every month, with DataForSEO metrics. Twenty-five are free with no key; the free plan builds ten links on them, one of which may be on a DA 90+ site, and the rest appear locked by name and DA.
  • It is honest about limits. A captcha or a login is a gate it asks you about, never something it fakes its way past. Links that are plain text, nofollow or on a noindex page are reported as such.

Gates are handled in the chat, not skipped

A captcha, an email verification, a phone code, a social login or a payment step is a gate. The agent stops at the gate, tells you what the site wants, and offers the ways through, in the same conversation:

  • Clear it yourself: it opens the page in a visible browser window and waits while you solve the captcha or log in.
  • Connect a service once: a captcha solver, your inbox over IMAP, or an SMS number. From then on the agent reads the verification link or code itself and continues without asking.
  • Paste the code: you read the email or SMS and paste the code into the chat.
  • Skip the site: it moves to the next site in the plan.

It never fakes a person or works around a site's rules, which keeps your site out of trouble and the links worth having.

Everything a subscription adds

Activate in the chat; nothing else changes.

Feature What you get
Model-driven executor Builds on account-based sites: sign-up, profile fields, articles, forum posts, directory listings, page builders
Your own AI model Connect an Anthropic, OpenAI, Google or OpenRouter key once; it drives the browser on account-based sites, so model cost is yours and under your control
Gates and services Captcha solver, IMAP inbox and SMS connected once; verification links and codes read automatically
Background jobs queue_build starts a build and job_status polls it, so clients with short tool timeouts never drop a link
Identity generation generate_identity creates a complete, consistent sign-up identity on your catch-all domain
Monitoring recheck_links re-fetches every placed link and marks the ones that dropped
Campaign reports campaign_report gives planned versus placed, counts by status and method, and every live URL; CSV export
Dashboard A web page per API key with credits, campaigns, results and proof screenshots
Competitor gap The referring domains your competitors have and you do not, each row carrying the method to build it
Keyword and anchor suggestions Read a target page and get keywords and a full anchor plan across the five types
Link velocity pacing Links spread across days and methods, never fired in a burst
Multi-site Register many sites under one subscription, each with its own caps and reports
Fair use 500 placed links per key in any 30 days; only verified live links count

Configuration

Nothing is required. activate stores your key in ~/.seoagent/config.json; results live in ~/.seoagent/results.db. Two optional variables override that:

Variable Purpose
SEOAGENT_API_KEY A subscription key, for environments where a config file is inconvenient (CI, containers).
SEOAGENT_URL The hosted endpoint, or a personal link https://mcp.seoagents.dev/u/le_…/mcp. Only needed for a self-hosted engine.

Where it runs

One engine, one library, one subscription. Four ways in, at four different stages.

Door Status What it is
Any AI assistant Live Claude Code, Claude Desktop, claude.ai, Cursor, Codex, ChatGPT, Windsurf. pip install seoagents, ten free links, built in your own browser.
WordPress plugin In development Posts, pages and WooCommerce products as targets. Campaigns, gates, weekly monitoring and white-label reports inside wp-admin.
Shopify app In development Products and collections as targets, embedded in the Shopify admin, billed through Shopify.
Windows and macOS app In development The only door that uses accounts you are already signed in to. Its own browser, your own model key, nothing uploaded.

The roadmap says plainly what is live, what is being built, and what will never be built.

Guides

Documentation site: https://m4mansoor.github.io/seo-agent/

Licence and author

MIT. Built by Engr. Inaamul Haq Mansoor (@m4mansoor).

Metadata

Release files for seoagents 0.2.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for seoagents 0.2.0
File Size Uploaded
seoagents-0.2.0.tar.gz 701.3 kB Details

Built distribution (wheel)

Table of built distributions (wheels) for seoagents 0.2.0
File Interpreter ABI Platform
seoagents-0.2.0-py3-none-any.whl Python 3 none any Details

Total release size: 782.3 kB

Release files / seoagents-0.2.0.tar.gz

Download URL seoagents-0.2.0.tar.gz
Size 701.3 kB
Tags Source
SHA-256 checksum
How to use checksums
0f26fbeeac244507338d10da1975f216e37dd79f5b1ad95f16d88d1d78ab17cb
BLAKE2b-256 checksum
How to use checksums
8841e5b21905baefd4e2875ec5926739099928feff85eb585df6e61d7e6de8f9
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.12.13 {"installer":{"name":"uv","version":"0.12.13","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

Release files / seoagents-0.2.0-py3-none-any.whl

Download URL seoagents-0.2.0-py3-none-any.whl
Size 81.0 kB
Tags Python 3
SHA-256 checksum
How to use checksums
085ce7eff749956993e07f68f2ce668c437c0b9ab635fa6dffac165a5505cfcd
BLAKE2b-256 checksum
How to use checksums
3115d4f1f33e49009c81316bf203e432ab440642e790195b4ab76223650204c0
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.12.13 {"installer":{"name":"uv","version":"0.12.13","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

Release history Release notifications | RSS feed

This release

0.2.0 This release

2 release files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page