Saylent

The Claude Code plugin

Two commands install Saylent in Claude Code - a free crawler check, an audit that shows its price before it spends, fix verification and report summaries.

The plugin puts Saylent inside Claude Code as six skills and the MCP server. Ask in plain words ("can AI crawlers read example.com?", "what do the assistants say about us?") or call a skill by name. It is the same pipeline as the CLI: your own keys, the same cost estimate before a run, the same reports on disk.

Install

In a Claude Code session:

/plugin marketplace add yotambraun/saylent
/plugin install saylent@saylent

Claude Code loads the plugin in your next session, or right away after /reload-plugins. The plugin starts the server with npx -y saylent@0 mcp, so it needs Node 22 or newer and nothing else installed. Fixes released on the 0.x line reach you without reinstalling.

The six skills

SkillWhat it doesCost
/saylent:ai-visibility-auditThe whole loop: a free estimate, your yes, the audit, the sites to get onto, the drafted fixes with their evidence (and saylent fixes apply for the site ones), and a verify planYour own credits for the audit only, shown before you agree; every other step $0
/saylent:gate-checkWhich AI crawlers the site lets in: robots.txt per bot class, a live request as each AI user agent, JSON-LD and meta directives, with the fix for each problem$0, no keys
/saylent:auditWhat ChatGPT, Claude, Gemini and Perplexity tell buyers: recommended or not, who is named instead, which pages are cited, and a fix plan. Writes run.json, report.html and report.mdYour own credits, shown before you agree
/saylent:verifyRe-asks the same frozen questions after you ship fixes and shows what moved, in movement.htmlYour own credits, about the same as the audit
/saylent:read-reportSummarizes a run you already have, or answers a question about one section$0, offline
/saylent:crawlersWhich AI crawlers actually visit, from your own access log: bots, purposes, pages, blocks and errors with redacted log lines, and with a run, the cited pages no AI crawler fetched$0, offline; the log stays on your machine

Each skill takes its input after the name, for example /saylent:gate-check example.com or /saylent:verify ./example.com/2026-10-01/run.json.

Start with gate-check. A blocked crawler explains most poor results, and it costs nothing to find out.

Spending is always your call

audit never spends first. It runs a free dry run, then tells you which provider keys it found, which engines it will ask, which model judges the answers, and the estimated cost, and asks before it runs. The run it starts carries a max_usd ceiling: if the estimate is higher, or the run would take today's spend past the daily cap ($20 by default, set with DAILY_SPEND_CAP_USD), the server refuses before anything reaches a provider. verify states the cost of the original run and asks the same way.

As a guide, with four engines a smoke audit (6 questions) costs about $0.80 to $1.20 and a full audit (23 questions) about $5.10 to $7.10. Fewer engines cost less. See What it costs.

Keys

gate-check, read-report and crawlers need no keys. audit and verify need one of OpenAI or Anthropic (both give cross-family judging); Gemini and Perplexity add two more engines, and a SerpApi key adds the answer surfaces. Set them up once, in any of these ways:

  • Recommended: run npx saylent keys set openai (or anthropic, gemini, perplexity, serpapi) in your own terminal. The key is typed into a masked prompt and saved to ~/.saylent/config.json with owner-only permissions, and the plugin finds it however Claude Code was started. npx saylent keys test checks each key with a free call.
  • Export OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, PERPLEXITY_API_KEY or SERPAPI_API_KEY in the shell you start Claude Code from. The plugin passes exactly these five to the server.
  • A .env file in the project folder, kept out of git.

Never paste a key into the chat. Keys are never tool arguments, and the skills report only which providers have a key, never a value.

Where the files go

Reports are written to ./<domain>/<YYYY-MM-DD>/ under the folder Claude Code runs in, so read-report and verify find them later. The server refuses an output folder outside that working directory, and the other MCP safeguards apply unchanged.

Updating and removing

Run /plugin and open the Installed tab to update or uninstall, or run claude plugin update saylent@saylent in your shell.

Without the plugin

The plugin is a convenience over the MCP server. To add only the server, with no skills:

claude mcp add --transport stdio --scope user saylent -- npx -y saylent mcp

The MCP server page has the tool reference and the setup for Claude Desktop, Cursor, VS Code, Windsurf, Codex CLI and Gemini CLI.

On this page