Skip to main content
Glama
site-context-mcp
by YongBoYu1

site-context-mcp

Problem

Agents invent project and site claims without citations. When you ask what a public site says about a product page, a topic note, or an essay, you often get fluent prose and no way to verify it against the live corpus.

Related MCP server: llms-txt

Solution

A stdio MCP server that:

  1. Fetches public pages starting from llms.txt, then follows same-host links to core pages

  2. Caches them in-process (~1-hour TTL)

  3. Exposes tools that return structured answers with a top-level sources[] of public URLs on every response

No secrets, no LinkedIn scraping, no private admin / WOD-APP links — only HTTPS to the wired public site host.

This package is for site/project context. It does not duplicate resume identity or resume-summary tools (see sibling resume-mcp for that).

The default corpus is YongBo Yu’s public site. The same fetch/cache/source-URL shape can be pointed at any public llms.txt-indexed site corpus.

Requirements

  • Python 3.11+

  • Network access to the wired public host (default: yongbo-yu.vercel.app) when running the live server

Install

# from this repo
pip install -e .

# or with uv
uv pip install -e .

Dev / smoke tests (mocked HTTP; no live network required):

pip install -e ".[dev]"
pytest -q

Run (stdio)

# module entrypoint
python -m site_context_mcp

# console script (after install)
site-context-mcp

# with uv
uv run python -m site_context_mcp

The process speaks MCP over stdin/stdout. Do not pipe unrelated stdout into the same process.

Cursor mcp.json

Add a server entry (Cursor: Settings → MCP, or edit ~/.cursor/mcp.json / project .cursor/mcp.json):

{
  "mcpServers": {
    "site-context-mcp": {
      "command": "python3",
      "args": ["-m", "site_context_mcp"],
      "cwd": "/absolute/path/to/site-context-mcp"
    }
  }
}

If the package is installed into a venv, point command at that interpreter (or use the site-context-mcp console script).

With uv:

{
  "mcpServers": {
    "site-context-mcp": {
      "command": "uv",
      "args": ["run", "python", "-m", "site_context_mcp"],
      "cwd": "/absolute/path/to/site-context-mcp"
    }
  }
}

After saving, reload MCP servers in Cursor. The tools below should appear as site-context-mcp tools.

Tools

Tool

Purpose

list_site_pages

Catalog of cached public pages from llms.txt + linked core pages

get_page

One page by path (/projects/kilodock) or absolute HTTPS URL

search_site

Keyword search over the cached corpus (snippets + source URLs)

get_project

Project evidence page (default: kilodock), with optional light resume.json cross-link

Every tool response is JSON text that includes a top-level sources array of public URLs. Partial fetch failures are reported in cache_notes when present.

Example prompts (in Cursor / Claude)

  • “List the public pages this MCP has cached and cite the source URLs.”

  • “Get /projects/kilodock and quote the engineering metrics with sources.”

  • “Search the site corpus for ‘LangGraph’ / ‘Codex’ and return snippets with URLs.”

  • “Use get_project for KiloDock evidence — do not invent stack claims.”

Data sources (public only)

On startup (and on first tool use if needed), the server fetches:

  1. https://yongbo-yu.vercel.app/llms.txt

  2. Same-host pages linked from llms.txt (home, about, project pages, topic pages, essays)

  3. Optionally resume.json when linked — used only for light project cross-links, not as a resume clone

Binaries and off-host / private targets are skipped. No API keys.

Author

YongBo Yu (also Yong Yu) — Toronto, Canada · GitHub YongBoYu1

License

MIT — see LICENSE.

Available Tools

4 tools
get_pageA

Fetch one cached public page by path (e.g. /about-yongbo-yu) or absolute HTTPS URL.

Inputs: path_or_url (required). Returns: result

get_projectB

Return public project evidence for a site project page (default: kilodock).

Inputs: name_or_slug. Returns: result

list_site_pagesB

List cached public site pages discovered from llms.txt (paths, titles, URLs).

Inputs: no named parameters. Returns: result

search_siteB

Keyword search over the cached site corpus; returns snippets with source URLs.

Inputs: query (required), limit. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.8/5.0

    Scored across 4 tools

    Disambiguation5/5

    Each tool has a clear, non-overlapping purpose: listing pages, fetching a single page, searching the corpus, and retrieving project evidence. No two tools could be confused for the same operation.

    Naming Consistency5/5

    All tools follow a consistent verb_noun pattern in snake_case: list_site_pages, get_page, search_site, get_project. The naming is uniform and predictable.

    Tool Count5/5

    Four tools is a well-scoped set for a site-context server—enough to cover enumeration, retrieval, search, and specialized project data without redundancy or bloat.

    Completeness5/5

    The surface covers the core needs for browsing a cached site corpus: list, get, search, and project-specific lookup. There are no obvious missing operations for a read-only context server.

    Maintenance

    ActivityStale
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      Not graded
      maintenance
      Enables querying and retrieving content from webpages by parsing sitemap.xml files and fetching HTML content from specified URLs. Includes rate limiting for abuse protection.
      -
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables agents to discover the llms.txt index covering any site, fetch its linked documentation pages as markdown, and search across parsed indexes by section heading, link title, and notes.
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables AI agents to perform LLM-optimized web searches, retrieve recent news articles with full text, and extract clean readable content from URLs. It returns synthesized answers, ranked cited sources, and page text suitable for grounding RAG workflows.
      310 npm
      MIT
    • A
      license
      Not graded
      quality
      A
      maintenance
      Fetches webpages and converts them into clean, token-efficient Markdown, optionally crawling multiple linked pages while respecting robots.txt and caching results for AI agents and LLM pipelines.
      MIT