The SEO MCP server your AI agent should use depends on the job. Investigating a broken page on your own site needs different evidence from researching a competitor's backlink profile.
I'd start with the service that already holds the data you need, then check its account requirements, usage limits, and supported actions. Connecting five servers won't automatically produce a better answer.
These are five official SEO MCP options, compared using their documentation checked on October 27, 2026. I'm a co-founder of Screpy, which is included. This is a researched comparison, not a hands-on accuracy benchmark. The product images come from each site's homepage social preview metadata.
Compare access and limits before tool counts
MCP stands for Model Context Protocol. It lets a compatible AI client discover and call tools provided by a server. In SEO, that can mean retrieving a crawl finding, querying keyword data, or inspecting a saved ranking history.
The useful comparison is what your account can actually retrieve.
| Server | I'd consider it for | Access and cost model | Limit worth checking |
|---|---|---|---|
| Screpy | Investigating managed projects and their SEO evidence | Screpy subscription; regular Pro is $192/year | Project capacity and feature eligibility; Pro excludes AI Visibility |
| Ahrefs | Keyword, competitor, and backlink research | Paid Lite plan or higher; plan API allowance | Lite: 100 rows per request, 200,000 API units/month |
| Semrush | Research using an eligible Semrush API subscription | Eligible plan/API units; Traffic & Market data needs Trends access | Units consumed by report and amount of data |
| SE Ranking | SEO/GEO research with subscription credits or separate API access | Included API credits, or pay-as-you-go from $50 for 250,000 credits | Subscription access and available credits |
| DataForSEO | Workflows built around specific data API requests | Pay-as-you-go; $50 minimum balance top-up | Endpoint, request parameters, and repeated calls |
These aren't equivalent allowances. An API unit isn't necessarily one request, and one prompt can lead to multiple tool calls. Your AI client's own fees and limits also sit outside this table. Follow the linked official sources below before buying or connecting anything.
1. Screpy MCP: investigate the projects you manage
Screpy MCP makes sense when your question starts with a website you manage: which pages have broken links, what changed in Search Console, or what your saved tracking results show.
The hosted server uses browser-based OAuth. The setup documentation gives the address https://mcp.screpy.com and recommends listing your accessible projects first. It requires a client with remote HTTP MCP and OAuth support; a Screpy REST API key isn't the sign-in method.
A useful first task would be to inspect a completed crawl, find a broken internal link, and identify the page containing it. That gives a coding agent evidence it can use while reviewing the relevant template. Screpy MCP itself doesn't repair your website's markup or publish the fix.
Keep the data boundary clear. Saved rankings aren't a live SERP lookup, and this connection isn't a general keyword-discovery service. Stored AI Visibility answers also need their analysis dates, prompts, and citations attached.
On Screpy's pricing page, Pro's regular annual price is $192, paid upfront for the year. It includes five projects, with 5,000 crawl pages and 100 tracked keywords per project. AI Visibility and backlinks are excluded from Pro. Max's regular annual price is $456 and includes those features, with 50 AI Visibility prompts per project.
Welcome offers can change. Check the final total and renewal conditions; subscriptions renew unless cancelled. MCP access doesn't unlock features or extra capacity outside your plan.
2. Ahrefs MCP: research competitors, keywords, and backlinks
Ahrefs SEO MCP is a candidate when you need external research: keyword opportunities, competing domains, and backlink profiles.
For example, you could ask an agent to compare two competitors and return a small set of relevant pages or referring domains to inspect. Traffic estimates and authority metrics would still be research signals, not access to either company's analytics or proof of sales.
MCP is available from Lite upward. Ahrefs lists a Lite limit of 100 rows per request and 200,000 monthly API units. That is not 200,000 free prompts or unlimited exports. Check the localized plan pricing and the allowance that applies to your workspace.
The official MCP documentation describes hosted access and authorization. OAuth-created keys also have a default monthly limit, which can constrain a connection separately from the workspace allowance.
There's a developer-specific boundary worth reading: Ahrefs says its external MCP endpoint isn't a general-purpose API for custom scripts, bridges, or standalone HTTP clients. Use its supported public API for programmatic integration. An MCP connection inside a supported assistant and an unattended custom data service are different use cases.
3. Semrush MCP: use the API access your plan includes
Semrush MCP is worth checking before adding another subscription if your team already has eligible Semrush access.
Its current developer documentation lists SEO Classic Pro and Guru, plus Semrush One Starter and Pro+, with 50,000 API units that refresh on renewal. Other listed options, including SEO Classic Business and Semrush One Advanced, require an API unit package. Traffic & Market data requires a Trends API subscription.
The connection supports OAuth, with API-key authentication available for compatible clients that don't support OAuth. There isn't an additional MCP usage fee, but report requests consume API units.
An important distinction: project methods exposed through this MCP are read-only. Reading a project's audit or tracking data doesn't mean the assistant can create projects or modify their settings.
I'd choose it when its research reports answer the question and the existing account allowance is sufficient. Confirm the report you need is included before building a recurring agent workflow around it.
4. SE Ranking MCP: choose subscription credits or API access
SE Ranking MCP brings keyword, domain, backlink, audit, and AI search visibility data into a hosted connection.
Its access model is useful to compare with a new full-platform purchase. MCP runs on the API credits included in a subscription, but there is also a standalone pay-as-you-go route advertised from $50 for 250,000 credits. The trial advertises 100,000 credits. Those are data credits, not a promise of that many completed agent tasks.
The setup documentation describes Streamable HTTP and OAuth at https://api.seranking.com/mcp. Available tools still depend on the subscription.
Check account ownership if you're working in an agency workspace. The docs flag a sub-account limitation: API access belongs to the master account, so a sub-account's normal sign-in may not expose tools. Have the administrator review the supported connection route and access before assuming login alone is enough.
I'd shortlist it when the research coverage fits and you want to evaluate API-only spending alongside a platform subscription. Keep the question small enough to understand what consumes the credits.
5. DataForSEO MCP: build around specific API requests
DataForSEO MCP is a candidate when you want an agent to work with specific SEO data APIs rather than start from a managed project's dashboard.
Its API categories include SERP data, keywords, backlinks, and on-page analysis. The official server repository documents hosted remote access with OAuth, plus local HTTP and stdio options.
The current implementation exposes documentation-discovery tools and an api_request tool. That makes request selection important: have the agent inspect the relevant documentation, choose the endpoint and parameters, then request only the data needed. Don't assume every call has the same price or response size.
Pricing is pay-as-you-go, with a $50 minimum top-up. That money becomes account balance; it isn't a $50 monthly subscription. Unused balance doesn't expire, and actual charges depend on the API service and request.
I'd consider it for a workflow where you want to control the data query closely. That flexibility also puts more responsibility on the workflow: repeated requests, larger responses, and unnecessary follow-up calls can change what a task costs.
Run one bounded task before committing
I'd make the first decision with a small acceptance check, not the number of tools in a server's catalog.
Start with this prompt in your chosen assistant:
Show which relevant data and read-only tools this connection makes available to my account. Tell me the plan or usage constraints you can verify. Don't run a crawl, add tracking, change settings, or edit a website.
Then choose a task that fits the answer. For a managed project, inspect one completed crawl and return a few affected URLs. For a research connection, investigate one keyword and two competitor domains in a specified market. Neither example is a claim that all five servers support both jobs.
Ask for a compact result with:
- The exact project, domain, keyword, and market used.
- The source report and its date or reporting period.
- Evidence behind each recommendation, with missing data identified.
- Usage consumed, if the service exposes it; otherwise, an explicit unavailable value.
- A clear statement of whether anything was changed.
Stop if the required report isn't available. A fluent answer built from a different dataset doesn't solve the missing-data problem. Likewise, an empty result doesn't establish that your site has no issues.
If you're ready to turn the evidence into a workflow, my SEO tasks you can automate with ChatGPT article covers practical next steps.
Choose the connection that can support your actual decision with traceable evidence. Add another server when it fills a specific gap, rather than because it adds more tools to the menu.
Top comments (0)