Agentready MCP

What AshutoshRaj97/agentready-mcp MCP server does

AshutoshRaj97/agentready-mcp MCP server connects an MCP-compatible agent to AgentReady, a hosted service for indexing and querying public websites. An agent can check which domains are already available, submit a new site, and ask natural-language questions about its indexed content. Answers are grounded in the crawled pages and include citations to source URLs.

The service is aimed at documentation and other public web content. It can work with static sites, server-rendered pages, and JavaScript-heavy applications such as React, Next.js, Vue, Docusaurus, and GitBook sites. It does not index sites that require authentication or provide no usable public HTML content.

How it works

When a site is submitted, AgentReady first looks for llms.txt or llms-full.txt. If those are unavailable, it performs an HTML crawl using a sitemap or discovered links, extracting up to 10 pages. Next.js data embedded in __NEXT_DATA__ can provide another source, and a Jina Reader rendering fallback handles pages whose useful content appears only after JavaScript runs.

Indexing usually takes about 60 seconds. Once complete, ask_site can synthesize information from multiple pages without requiring the agent to fetch each page separately. A new domain can also be indexed automatically when an agent asks a question, checks capabilities, or creates a plan.

AshutoshRaj97/agentready-mcp MCP server does not execute actions against the indexed website. Its planning feature returns supported steps, evidence, risks, confirmation requirements, and a receipt for a future operation, while remaining read-only. Use refresh_site when content has changed or an answer appears stale.

Setup and configuration

Run the local MCP bridge with:

npx -y @agentreadyweb/mcp

MCP clients using stdio configure npx as the command and pass -y and @agentreadyweb/mcp as arguments. The README gives configurations for Cursor, VS Code agent mode, Windsurf, and other stdio-compatible clients. A client supporting HTTP transport can instead connect to https://www.agentready.it.com/api/mcp without installing the npm package.

No account is required to index the first site. The material does not specify environment variables or API credentials for the MCP bridge.

Tools and capabilities

The available operations include:

• list_sites: view indexed domains, titles, and page counts.
• get_site_capabilities: inspect freshness, schemas, endpoints, and read-only limits.
• get_site_facts: retrieve structured facts such as pricing tiers, features, and limits.
• plan_site_action: produce an evidence-backed, non-executing plan.
• submit_site: index a public URL.
• ask_site: ask questions and receive cited answers, with optional URL scoping.
• refresh_site: perform a full re-crawl.
• rate_answer: submit a quality score from 1 to 5, optionally linked to a request ID.

The same package also provides CLI commands for grading a site's agent readiness, asking questions, indexing, and refreshing. JSON output is available for CLI automation, and the grading command can be used as a CI check.

Limitations and notes

AshutoshRaj97/agentready-mcp MCP server depends on AgentReady's hosted indexing and retrieval service. Crawling is not instantaneous, particularly for a new or refreshed domain. Sites behind authentication or without public HTML content are outside the documented indexing scope.

The plan operation is deliberately read-only and does not perform side effects. The README also documents a deploy webhook for triggering refreshes; it is rate-limited to once per hour per domain and requires no authentication. For Docusaurus, MkDocs, Starlight, or Sphinx documentation sites, separate plugins or extensions can trigger indexing during builds, but these are additional integrations rather than part of the MCP bridge itself.