Web Speed

Web Speed is a deterministic translation layer built for agentic AI systems, converting messy human-facing web pages into precise, structured JSON maps. Developed under the brand Webspeed Infrastructure, it is currently at System Version 2.0.4 and under active development. The tool targets the core inefficiency of AI agents consuming raw DOM trees — a process that wastes tokens and introduces hallucination risk — by stripping out HTML noise and delivering clean structural blueprints instead.

What It Is

Web Speed sits between the raw web and an AI agent's context window, acting as an infrastructure-level "adaptation layer." Rather than letting an agent parse hundreds of kilobytes of raw HTML, Web Speed fetches a page, extracts its structural map (navigation, forms, links, text blocks, headings), and returns a compact JSON object. The docs describe token savings of typically 75–97% per page fetch. The tool is available as both a self-hosted open-source MCP server and a managed hosted API.

How the Extraction Engine Works

Every Web Speed response follows a standardized JSON schema regardless of the target site's styling or layout changes. The schema includes:

• headings — structured by level
• navigation — labeled links with location context (header, footer, etc.)
• forms — field names, types, and placeholders
• content_links — total count and labeled items
• text_blocks — tagged paragraph content
• metadata — language and page-level attributes

This deterministic output means agent logic doesn't break when a site redesigns its CSS. The about page frames this as a "division of labour": Web Speed handles fetching, extraction, caching, and form submission; the agent handles goal-setting, planning, and judgment.

MCP-Native Session Handling

A notable feature highlighted on the homepage is the Native MCP integration for post-authentication workflows. Web Speed uses a local browser bridge to inject session tokens from the user's local keychain, enabling agents to operate behind login walls without routing credentials through a proxy. The homepage describes this as a "New Launch" capability, positioning it for use cases like navigating LinkedIn Jobs or other walled-garden platforms.

Deployment Model: Open Core

Web Speed follows an open-core model. The community edition is a self-hosted MCP server published on GitHub under the handle Dominic-Pi-Sunyer/web-speed-oss. It includes full JavaScript rendering, all 8 MCP tools, and a local 24-hour disk cache. The hosted API adds a shared map registry, priority support, and unlimited fetches at scale. The about page describes an "Open Core Flywheel" strategy, with the goal of partnering with agentic frameworks such as LangChain, CrewAI, and LlamaIndex to position Web Speed as a reliability layer across the agent ecosystem.

Update: System Version 2.0.4

The homepage displays "System Version 2.0.4 - Active" and the docs page notes "Active Development -- frequent updates," signaling ongoing iteration. The post-auth MCP session feature is labeled "New Launch" on the homepage, indicating it is a recently shipped capability. The project's GitHub repository serves as the primary distribution channel for the open-source edition.