Agent access over MCP and REST

SEO, AEO, and GEO tools for your AI agent

Give Claude Code, Codex, Cursor, or your own agent all 51 tools through one MCP server or REST API. The agent checks your live site or a preview deploy, gets a prioritized list of changes, and makes them in your codebase. Runs cost the same credits as in the browser.

Create an API key

Assessments your agent runs appear in your reports with the same scores, so you can follow the progress in the browser. You can also check your website there yourself.

How it works

  1. Step 1

    Create an API key

    Sign in and open API keys. The key is shown once and spends your workspace's credits.

  2. Step 2

    Connect your agent

    Paste the config for your client below and replace YOUR_API_KEY. Any MCP client that supports Streamable HTTP with headers works.

  3. Step 3

    Ask for a check

    Ask in plain language, such as "check example.com and fix what you find". The agent runs one website assessment and applies the fixes in your code. After you deploy, it assesses again to show the score change and the share of issues fixed.

What your agent runs

The skill and the MCP server instructions steer your agent to this plan. Before running more than the default assessment, it confirms the credit total with you.

One website assessment

67 credits for one page, 47 for each extra page

In one call, the run-assessment tool runs Robots.txt & AI Crawler Checker, Redirects, Headers & Host Checker once per site. It runs Indexing & Canonical Checker, On-Page SEO Checker, Schema Markup Checker, Readability Analyzer, Citability Analyzer, Content Freshness Checker, Author & E-E-A-T Checker on each key page. It returns SEO readiness, AEO readiness and AI access scores from 0 to 100 and one fix list.

The next assessment of the same URL reports the percent of issues fixed, with the scores before and after.

Checks that fit the site

Added only when the condition holds

AI answer sampling

Only when you ask, with the questions agreed first

These sample live AI answers, which vary between runs, so treat them as observations to track, not fixes to verify.

Improvement tips

Every successful run returns improvementTips next to the result. They say what to change, where, why it matters, and how to verify the fix. The browser and stored runs show the same tips.

status
changes when there are tips; clean when the tool checked everything and found nothing to change; incomplete when part of the check could not run.
summary
A plain-text checklist of every change, ordered by priority.
tips
One entry per change with a priority, the action, and the affected targets. Ids stay the same across runs, so a fixed tip is recognized by its absence.
unchecked
The parts of the check that could not run, such as a data provider outage, and what to do about each.
rerunVerifies
False for generators and pasted input, whose tips come back for the same input. Apply those once.
verify
How to confirm the change worked: deploy, then run the same tool with the same input.
fixIn
dns or external when the change is made outside your code, so the agent tells you instead of editing files.

Your agent fixes what lives in your code and tells you about DNS, CDN or hosting changes it cannot make. A check marked incomplete stays unverified until it runs in full. Brand Narrative differences need your review before anything on the site changes.

Connect an MCP client

MCP endpoint https://seogeoaeo.ai/api/mcp. Streamable HTTP, stateless, with Authorization: Bearer on every request.

Claude Code

Run once in a terminal. The server is added for your user and every project.

Cursor and Windsurf

Add to ~/.cursor/mcp.json, or to Windsurf's mcp_config.json.

GitHub Copilot in VS Code

Add to .vscode/mcp.json in your workspace, then start the server from the file.

Codex

Add to ~/.codex/config.toml and export SEOGEOAEO_API_KEY in your shell.

Claude Desktop

Add to claude_desktop_config.json. Claude Desktop starts local servers from this file, so it connects through mcp-remote, which needs Node.js.

Gemini CLI, Cline, OpenCode, and other clients

Any client that supports remote Streamable HTTP servers with headers works. Use the endpoint https://seogeoaeo.ai/api/mcp and the header Authorization: Bearer YOUR_API_KEY. Only the name of each config key differs between clients.

Agent skill

A SKILL.md file that tells your agent which tool fits a request, what it costs, and how to call it. Works with Claude Code and any agent that loads skills.

Or open SKILL.md and save it in your agent's skills folder.

REST API

Base URL https://seogeoaeo.ai/api/v1. Every tool has a typed request body in the OpenAPI document, so you can generate a client or import it as a GPT action.

Allow four minutes for AI answer, brand narrative and fan-out reports. Some providers take over two minutes to answer. If a request times out, retry with the same idempotencyKey. Check improvementTips.unchecked for any provider data that could not be retrieved.

maxCredits
Refuses the run with a 402 if the price is above this number.
Assessments
POST /assessments runs checks for 40 seconds, or as long as the slowest selected check (up to about three and a half minutes). While status is in_progress, call POST /assessments/{id}/continue.
idempotencyKey
Send the same key on a retry to get the original run instead of paying again. A key is remembered while its run is among the workspace's latest 1,000.
Errors
Always { "error": { "code", "message" } }, with a stable code such as insufficient_credits or rate_limited.
Free calls
Listing tools, reading runs and assessments, and checking credits cost nothing. Over MCP they are check-credits, get-run and get-assessment.

Full reference for agents and LLMs: llms-full.txt.

Tools your agent gets

Each tool is an MCP tool and a REST path named by its slug, next to run-assessment and get-assessment for whole-site assessments.

SEO

Technical checks, on-page analysis, keywords, and Core Web Vitals.

AEO

Structured data, entity markup, and answer formatting. Content checks also support answer readiness.

GEO

AI access, content signals, sampled brand mentions and citations. Optional agent utilities are separate from search requirements.

Questions

How long can AI reports take?
Allow four minutes for AI answer, brand narrative and fan-out reports. Some providers take over two minutes to answer. If a request times out, retry with the same idempotencyKey. Check improvementTips.unchecked for any provider data that could not be retrieved.
Can AI agents use seogeoaeo.ai tools?
Yes. Every tool is available over an MCP server at https://seogeoaeo.ai/api/mcp and a REST API at https://seogeoaeo.ai/api/v1. Any MCP client can connect, including Claude Code, Cursor, VS Code, Codex, Windsurf and Claude Desktop. It sends an API key as a bearer token.
How do agents authenticate?
Create an API key at https://seogeoaeo.ai/api-keys. Send it on every request as "Authorization: Bearer <key>". Keys start with sga_, are shown once, and can be revoked at any time.
Can a coding agent fix what the tools find?
Yes. Every successful run returns improvementTips. They hold a status, a plain-text summary and a list of changes ordered by priority. Each change names the URLs or items it applies to. The tips also list any parts of the analysis that could not run, and how to verify the fix. An agent working in the site's codebase applies the tips, deploys, and runs the same tool again until the status is clean. Tips keep the same id across runs, so a fixed tip is recognized by its absence. Generators and runs on pasted input return the same tips for the same input. They set rerunVerifies to false, so the agent applies their tips once.
How does an agent show that its fixes worked?
It runs a new website assessment of the same URL after deploying. Each assessment is compared with the previous one automatically. progress.percentFixed is the share of the earlier issues that are gone. Next to it are the fixed, still present and new counts, and the scores before and after. Issues are counted per page and item, so a fix applied to two of five pages shows as partly fixed instead of unchanged. Tips marked fixIn need a change in DNS or another service, which the agent hands to you.
What does an agent run cost?
The same credits as a run from the browser. The core checks cost 67 credits for one page and 47 for each extra page. AI answer sampling costs 95 to 150 credits per run, so agents run it only when asked. Agents can pass maxCredits to refuse a run above an expected price, and a run that fails before producing a result is refunded.
How do I stop an agent from paying twice on a retry?
Pass the same idempotencyKey when retrying a run. The API returns the original run instead of charging again, as long as that run is still among the workspace's latest 1,000.
What is the seogeoaeo.ai agent skill?
A SKILL.md file at https://seogeoaeo.ai/skills/seogeoaeo/SKILL.md that tells an agent which tool fits a request, what each one costs, and how to call it. Save it in your agent's skills folder, for example ~/.claude/skills/seogeoaeo/SKILL.md for Claude Code.
Which sites can the tools check?
Public http and https URLs only. Private, local, and non-web addresses are rejected before any request is made.