# Run a technical SEO audit from Claude or ChatGPT

> Ask your agent to crawl your site and report what is broken, what Google cannot index and what to fix first. Manifold crawls in the background at 5 credits for each 100 pages asked for, and your agent reads the result when it is ready. Our 146-page site scored 97.93 out of 100, with nothing broken and five titles too long.

| | |
| --- | --- |
| Job | A technical SEO health check, ranked by what to fix |
| Tools | seo_run_technical_crawl, get_task |
| Credits | 25 for 500 pages asked for |
| Checks | Status codes, broken links, indexing, duplicates, on-page issues |
| Run on | manifoldmcp.com, 27 October 2026 |

## The prompt

We asked our own agent, Claude Code connected to Manifold, on 27 October 2026:

```text
Run a technical SEO audit of manifoldmcp.com: crawl up to 500 pages and tell me what is broken,
what Google cannot index, and what to fix first.
```

## The calls

| Step | Tool | Input | Credits |
| --- | --- | --- | --- |
| 1 | [`seo_run_technical_crawl`](https://www.manifoldmcp.com/docs/tools/seo#seo_run_technical_crawl) | manifoldmcp.com, 500 pages | 25 |
| 2 | [`get_task`](https://www.manifoldmcp.com/docs/tools#async-tasks) | the crawl's task id, until it finished | 0 |

A crawl is slow, so the first call returns a task id and the agent reads the result with [`get_task`](https://www.manifoldmcp.com/docs/tools#async-tasks) when it is ready. The charge is on the pages asked for, 5 credits for each 100, not on the pages found: we asked for 500 and the site had 146.

## What came back

| Check | Result |
| --- | ---: |
| Pages crawled | 146 |
| On-page score (0 to 100) | 97.93 |
| Pages answering 200 | 146 |
| Broken links | 0 |
| Broken resources | 0 |
| Pages Google cannot index | 0 |
| Duplicate titles | 0 |
| Duplicate descriptions | 0 |

Three checks flagged pages:

| Issue | Pages | Where |
| --- | ---: | --- |
| Render-blocking resources | 146 | Every page |
| Low content rate | 95 | Home, the blog index, topic pages and other listing pages |
| Title too long | 5 | Four answer pages and the use cases index |

## What the agent concluded

- **The crawl found nothing blocking indexing.** Every crawled page answered 200, nothing was broken, and no robots.txt rule, header or meta tag kept a page out. Whether Google has indexed a given page is a separate check, in Search Console.
- **The long titles are ours to fix.** All five are answer pages or their index, where a title near our 60-character limit plus " | Manifold" runs past the crawler's limit. Shorter titles there are the first fix, and this page's own title is kept short.
- **Low content rate is mostly expected.** It means little text against the page's code. On listing pages made of cards, that is the design, not a fault. It matters only on a page meant to rank for a question.
- **Render-blocking resources are site-wide.** Every page shares one shell, so it is one speed fix there, not 146 page fixes.
- **We asked for more pages than we needed.** The site has 146 pages and the crawl charged for 500. A site this size can ask for 200.

## Run it on yours

```text
Run a technical SEO audit of {your domain}: crawl up to {pages} pages, tell me what is broken
or not indexable, and confirm the top issues on the pages that get traffic.
```

To check one page in full, call [`seo_get_page`](https://www.manifoldmcp.com/docs/tools/seo#seo_get_page). With Search Console connected, [`console_inspect_url`](https://www.manifoldmcp.com/docs/tools/console#console_inspect_url) tells you whether Google has indexed a page and why not, at no credit cost.

## Questions

### How many pages can the free plan crawl?

Up to 1,000 pages a crawl. The charge is on the pages you ask for, so ask for a little more than your site has.

### Does the crawl run JavaScript?

Not by default. Rendering is an option at ten times the cost, for sites that build their pages in the browser.

### Does it run Lighthouse or Core Web Vitals?

No. The crawl checks status codes, links, indexing, duplicates and on-page issues. It flags render-blocking resources but does not score speed.

### How long does a crawl take?

Minutes, depending on the site. Your agent starts it, gets a task id, and reads the result when it is done.

## Next

- [SEO MCP server](https://www.manifoldmcp.com/mcp/seo)
- [SEO tools in the docs](https://www.manifoldmcp.com/docs/tools/seo)
- [Google Search Console MCP server](https://www.manifoldmcp.com/mcp/search-console)

Published 27 October 2026. Updated 27 October 2026.