# Marketing and SEO teams: exports, crawls and web pages that drain your AI agent's tokens > Ads exports, analytics data and raw web pages fill your AI agent's context. Tokenade compacts them on your computer before the agent reads them. Source: https://tokenade.net/en/solutions/knowledge-work/marketing-seo · Updated: 2026-10-20 Marketing and SEO work with an AI agent means feeding it a lot of raw material: Ads and analytics exports, Search Console data, crawl results and dozens of web pages. Most of that material is markup and repeated rows, and the agent pays for every token of it. Tokenade compacts pages, exports and tool results on your computer before the agent reads them, so audits and reports cost less of your plan. This page is for marketers, SEO consultants and agency teams who already work with an AI agent that runs on their computer: Claude Code (including the Code tab of Claude Desktop), Codex, Cursor and similar. Sources below were read on 2026-10-15. ## What you deal with every day Marketers have adopted Claude Code quickly, and the workflows are data-heavy by design. Anthropic's own growth team uses it to produce ad variations exported "into a CSV file ready to upload directly into Google Ads" ([Anthropic, read 2026-10-15](https://claude.com/resources/articles/how-anthropic-uses-claude-marketing)). Four sources of token weight come up again and again. ### Weekly exports from every platform Client folders fill up with CSV files. One roundup of real marketing uses of Claude Code starts from people "pulling CSV exports manually from Google Ads, GA4, and Meta every week" ([Coupler.io, 2026-07-14](https://blog.coupler.io/claude-code-for-marketing/)). The same article warns that "teams managing multiple clients or running frequent data-intensive analyses may need the Max plan". ### Analytics connectors return a lot of rows SEO audits now plug the agent straight into the data. A French marketer describes connecting Claude Code to Google Analytics 4 and Search Console through two MCP servers, over twelve months, with every URL analysed individually ([Captain Marketing, 2026-05-05](https://blog.captainmarketing.io/audit-seo-claude-code/)). Each of those tool calls brings back long tables that stay in the conversation. ### Raw web pages are mostly markup Competitor pages and SERP research mean reading HTML full of navigation, scripts and tracking code. Cloudflare measured its own announcement post at 16,180 tokens in HTML against 3,150 tokens as markdown ([Cloudflare, 2026-02-12](https://blog.cloudflare.com/markdown-for-agents/)). Multiply by every page in an audit. ### Audits run many agents at once Popular SEO toolkits for Claude Code split the work across parallel agents. Claude SEO, for example, "runs 26 sub-skills and 19 specialist agents in parallel" ([claude-seo, 2026-02-07](https://github.com/AgriciDaniel/claude-seo)). Each agent reads its own pages and files, so the same weight is paid several times. ## What Tokenade changes for marketing and SEO work Tokenade sits between your AI agent and the pages, files and tools it uses. It changes what the model reads, not your data or your deliverables. - **Web pages come back as compact markdown.** `tokenade web` fetches a page without the markup, takes several URLs in one call, and with `--prompt "title tag, H1, internal links"` returns only the passages that answer ([web and search](https://tokenade.net/en/docs/web-search)). - **Search across several engines.** `tokenade search` runs one query across several search engines and returns titles, links and snippets, merged and ranked. - **Exports become compact tables.** CSV, JSON and `.xlsx` exports are compacted by format, and client reports in `.pptx` or `.pdf` are read as text ([documents and media](https://tokenade.net/en/docs/read-documents-media)). - **Connector results get folded.** Long results from MCP servers such as analytics or Search Console connectors are folded, with the full output kept on your computer if the agent needs it ([wrap MCP servers](https://tokenade.net/en/docs/mcp-wrap)). - **Long command output is folded.** Crawl scripts and other commands that print thousands of lines come back as a short summary. - **API keys stay out of the conversation.** Credentials Tokenade recognises, such as API keys and tokens in a script's output, are masked before the agent sees them. On our public benchmark, Tokenade makes long agent sessions [39% cheaper](https://tokenade.net/en/thol). A full site audit is a long session. ## What stays on your computer Compaction runs on the computer where you work. Your exports, client pages, prompts and usage counts stay on your machine ([What Tokenade sends and stores](https://tokenade.net/en/docs/privacy-telemetry)). The compacted text still goes to your AI provider, as it would without Tokenade. Tokenade masks credentials it recognises, but it does not remove client names, campaign data or figures from what the agent reads. > Let your agent read exports, analytics results and web pages as compact text, so every client audit costs less of your plan. > > Install: `npm install -g @tokenade/cli && tokenade install` — or let your agent do it: https://tokenade.net/install.md ## Setting it up You need an AI agent that runs on your computer. Claude Code, the Code tab of Claude Desktop, Codex and Cursor all work, and SEO skills you run inside Claude Code keep working. Claude Cowork is not supported, and in the Claude Desktop chat only MCP results are optimised ([compatibility](https://tokenade.net/en/docs/compatibility)). 1. Install the command line tool: `npm install -g @tokenade/cli`. 2. Run `tokenade install`. It finds your agents, connects to each one and wraps the MCP servers it can. 3. Run your next audit, then open `tokenade dashboard`. Full instructions are in [Install Tokenade](https://tokenade.net/en/docs/install). Tokenade is free. ## FAQ **Does Tokenade replace my crawler or SEO tool?** No. It does not crawl or score anything. It makes what your agent reads lighter: pages, exports, command output and connector results. Your crawler, Search Console and analytics stay the sources of truth. **Will it drop rows from my exports?** Exports are compacted by format, and long results are folded rather than deleted. The full output is kept on your computer, and the agent can get it back when a detail matters. **Can I keep my agent's own web search?** Yes. Set `TOKENADE_NO_WEBSEARCH=1` and run `tokenade install` once, and Tokenade leaves the agent's web search alone. Page fetching has its own switch, `TOKENADE_NO_WEBFETCH=1`. **Does it work with Claude Cowork?** No. Cowork runs in an isolated environment that local tools cannot reach. Tokenade works when an agent on your computer, such as Claude Code or Codex, does the reading. For other roles that work mostly with documents, see the [document work overview](https://tokenade.net/en/solutions/knowledge-work).