Sombra
Sign in
8 articlesShared 2 weeks agoLive

Sombra / About

What Sombra is and why it exists — product overview, core features (URL saving, collections, context distillation, MCP integration), and the research context: cognitive offloading, PKM, and the rise of context engineering.

Distilled Context

Sombra / About — Collection Context

What Sombra Is

Sombra is a research library for developers and AI agents. Tagline: "Your AI's research library."

Core positioning: not a bookmarking tool for human recall — a context infrastructure tool for AI agents. The workflow: Save web pages (Chrome extension) → Organise into Collections → Write distilled Collection Context → AI agents access everything via MCP.

MCP install command:

claude mcp add --transport http --scope user sombra https://sombra.so/mcp

Works with: Claude.ai, Claude Desktop, Claude Code, Cursor, any MCP-compatible client.

AI-Assisted Collection Building

Because Sombra is MCP-accessible, an AI assistant can autonomously build collections — researching a topic, saving relevant pages, organising them, and writing distilled context — all in a single conversation. The user describes what they need; the AI does the legwork.

Core Features

URL Saving & Web Clipping

  • One-click Chrome extension save
  • Clean markdown extraction (strips ads/nav/boilerplate)
  • Permanent archive — no dead links
  • Original URL and metadata preserved

Collections

  • Named, themed containers for saved content
  • Primary unit of context for AI agents
  • Organise by project, topic, or task

Collection Context / Distillation

  • Human-written or AI-written structured summary for a collection
  • Preserves code examples, API signatures, CLI commands verbatim
  • Signal-only, not a raw dump — optimised for AI attention budgets

MCP Integration

  • Remote MCP server at https://sombra.so/mcp
  • Connect once → all MCP clients access full library simultaneously
  • Authenticated per-user

Public Sharing

  • Share any collection via public URL
  • Public pages show: distilled context (full), notes (full), saved URLs as cited references (title + link + user annotation only)
  • Extracted page content stays private — public sharing surfaces your thinking, not other people's content

Use Cases

Developers / API research — Save docs for HubSpot UI components, Java Elasticsearch, AWS CloudFormation etc.; AI distils key interfaces and gotchas; every coding session starts with the agent already knowing the API.

Cybersecurity / vulnerability intelligence — Multiple CVE advisories, vendor patches, researcher writeups → single collection → distilled brief with affected versions, attack vectors, mitigations. Shareable as one unit.

Sales / prospect research — AI builds a collection on a prospect (website, press, job listings, funding) before a call; distils into a briefing.

Journalism / source collation — Save every source for a long-form story. AI surfaces contradictions, identifies gaps, drafts from a curated base.

Competitive intelligence — Living collection tracking a competitor's pricing, changelog, press, job postings. Always-current distilled brief.

Legal / compliance — Regulation, case references, guidance documents for a matter in one collection. AI drafts against citable primary sources.

Academic research — Papers, preprints, personal notes. Distilled literature review. AI writes with genuine grounding.

Writers / creators — Style guide, audience research, tone notes as a persistent AI briefing. AI knows your voice in every new conversation.

Intellectual Context

Cognitive Offloading

Clark & Chalmers (1998) The Extended Mind: cognition extends into the environment via tools. Risko & Gilbert (2016): cognitive offloading improves task performance at high cognitive load. Sombra externalises developer research so cognitive bandwidth goes to reasoning, not retention.

PKM (Personal Knowledge Management)

Discipline traces to Frand & Hixon (1999). Traditionally human-facing (Notion, Obsidian, Logseq optimise for human browsability). Sombra shifts the retrieval paradigm: collections organised for AI context injection. Distillation replaces search/graph traversal as the primary retrieval interface.

Context Engineering

Emerged 2025 as the central AI agent discipline. Anthropic: "the discipline of designing and building dynamic systems that provides the right information and tools, in the right format, at the right time." Core insight: most agent failures are context failures, not model failures.

Key data: context rot degrades performance well before technical limits (effective window often <256k tokens); sharded prompts cause 39% average performance drop. Raw saved pages injected wholesale are harmful context — distilled collection context is the correct pattern.

MCP as Infrastructure

  • Launched Anthropic Nov 2024, solving the "N×M integration problem"
  • Adopted by OpenAI, Google DeepMind, GitHub, AWS, Bloomberg, Cursor, VS Code within one year
  • 10,000+ published MCP servers by Dec 2025; donated to Linux Foundation AAIF Dec 2025

Competitive Differentiation

No one else does the full unified flow: save web pages → organise into projects → write distilled technical context → serve to coding agents via hosted MCP interface.

  • Notion/Obsidian/Logseq: human-facing, no MCP, no web extraction
  • Raindrop/Pocket/mymind: save-only, no distillation, no MCP
  • Will Larson's "datapacks": local files, manual management, single-client
  • Sombra: hosted, persistent, multi-client MCP-native, with distillation layer

Public sharing model is also differentiated: shares your research and thinking, not a mirror of third-party content.

Key References

  1. What is Sombra? — primary product overview (internal note)
  2. The Research Context: Why Sombra Exists — cognitive offloading, PKM, context engineering, MCP (internal note)
  3. Anthropic: Introducing MCP — https://www.anthropic.com/news/model-context-protocol
  4. Anthropic: Building Effective Agents — https://www.anthropic.com/research/building-effective-agents
  5. MCP Blog: One Year of MCP — https://blog.modelcontextprotocol.io/posts/2025-11-25-first-mcp-anniversary/
  6. Chroma Research: Context Rot — https://research.trychroma.com/context-rot
  7. Wikipedia: Personal Knowledge Management — https://en.wikipedia.org/wiki/Personal_knowledge_management
  8. PMC: Consequences of Cognitive Offloading (Grinschgl et al., 2021) — https://pmc.ncbi.nlm.nih.gov/articles/PMC8358584/