Intercept MCP

What bighippoman/intercept-mcp does

The bighippoman/intercept-mcp MCP server gives an MCP client tools for searching the web, fetching pages, researching a topic, extracting selected page values, and processing multiple URLs. Its main output format is Markdown, with headings, links, tables, code blocks, and other structure preserved where the selected fetcher supports it.

URL-specific handlers cover Twitter/X posts, YouTube videos and available transcripts, arXiv papers, text-layer PDFs, Wikipedia articles, and several GitHub URL types. GitHub handling includes repository README files, files, issues, pull requests, and releases. Direct image URLs in supported formats can be returned as MCP image blocks rather than text.

How it works

A fetch first checks the shared agentsweb.org cache and then uses site-specific handlers when the URL matches a known pattern. If those handlers do not produce content, the request moves through fallback strategies. These include browser rendering, Jina Reader, web archives, Common Crawl, native Markdown endpoints, direct fetching with Markdown conversion, RSS and scholarly or discussion sources, and Open Graph metadata. Some optional stages use services that require separate configuration, such as Cloudflare Browser Run, FlareSolverr, or a commercial web unlocker.

Results are also held in an in-memory cache. Successful and failed requests have separate time-to-live settings, and the cache uses a size limit with least-recently-used eviction. The fetch response reports details such as the source, quality, content length, truncation state, and the next offset for pagination.

Setup and configuration

Install bighippoman/intercept-mcp with the npm package through an MCP client that supports local stdio servers:

npx -y intercept-mcp

No API key is required for the core fetch tool. Optional configuration can improve coverage or search behavior. BRAVE_API_KEY enables Brave Search; otherwise the search tool can use SEARXNG_URL, then DuckDuckGo as a last fallback. GITHUB_TOKEN raises the unauthenticated GitHub API limit. Shared-cache behavior can be changed with INTERCEPT_SHARED_CACHE and INTERCEPT_CACHE_READ_ONLY.

Cache behavior is configurable with INTERCEPT_CACHE_TTL_MS, INTERCEPT_CACHE_FAILURE_TTL_MS, and INTERCEPT_CACHE_SIZE. These settings control successful-result lifetime, failed-result lifetime, and the maximum number of in-memory entries.

Tools and capabilities

• fetch retrieves one URL as Markdown, supports tier limits, length limits, pagination offsets, cache bypassing, and image blocks.
• fetch_batch processes up to 10 URLs in parallel with the same handler and fallback system.
• research searches for a query and fetches the top one to five results in the same call, with optional domain and freshness filters.
• search returns web results with query, domain, freshness, and page controls.
• extract returns selected values as JSON using CSS selectors, attributes, first-match or all-match behavior, and HTML-table conversion.

Limitations and notes

The bighippoman/intercept-mcp fetch result can be truncated by maxLength; callers should use the returned continuation offset with startIndex to retrieve later content. PDF extraction is specified for text-layer PDFs, not every kind of scanned document. Search freshness and pagination controls are ignored when the tool falls back to DuckDuckGo.

GitHub API requests work without authentication but are limited to 60 requests per hour; setting GITHUB_TOKEN increases that limit. Some fallback providers are optional and may need their own API token, paid account, or locally available service. Fetch quality and availability therefore depend on the URL, matching handler, cache state, and configured fallbacks.

Getting started with this bighippoman/intercept-mcp MCP server

Always refer to the official documentation for the most accurate and up-to-date information.