Open Source · MIT · 42 events / 278 edges in our dogfood graph

A CLI toolkit
for agentic AI.

Turn any MCP server — or RSS, markdown, URLs — into a typed knowledge graph. Query from your terminal, your agent, or over MCP. Your data stays local at ~/.contextix/graph.json.

$npx contextix ingest mcp ./hackernews-top.mjs

TypeScript · works offline (regex) or with ANTHROPIC_API_KEY (Haiku 4.5 extraction)

1 · Point it at your sources

$npx contextix ingest mcp ./hackernews-top.mjs

any MCP server → graph via a skill file (keyless)

$npx contextix ingest markdown ~/notes

your Obsidian vault → entities + wikilink graph

$npx contextix ingest rss https://news.ycombinator.com/rss

any RSS / Atom feed → typed events + relations

2 · Your agent queries the graph

$contextix why "AI export controls"

traces causal chain backward — influences / precedes / correlates

$contextix connect "Federal Reserve" "Bitcoin"

shortest BFS path between two entities across your graph

$contextix signals -t 7d --domain crypto

recent events filtered by window + domain

$contextix serve

expose the same graph as an MCP server (stdio)

Prefer MCP? Same graph, JSON-RPC interface

Claude Desktop · Cursor · any MCP-native client

.mcp.json
{
  "mcpServers": {
    "contextix": {
      "command": "npx",
      "args": ["contextix", "serve"]
    }
  }
}

5 verbs · CLI and MCP

why

Trace causal chain backward from any event.

$ contextix why "<event>"

mcp: contextix_why

connect

Shortest path between two entities or events.

$ contextix connect "<a>" "<b>"

mcp: contextix_connect

signals

Recent events with importance, confidence, tags.

$ contextix signals --domain crypto -t 7d

mcp: contextix_signals

entities

Search people, tokens, orgs, protocols, concepts.

$ contextix entities --search "fed"

mcp: contextix_entities

graph

Extract a raw subgraph for viz or analysis.

$ contextix export --format json

mcp: contextix_graph

Our dogfood — contextix running on live sources

live

We run the pipeline on curated crypto & AI sources every few hours. This is what the graph looks like in production — before you run it yourself.

Crypto5 recent
·

The Bifurcated Regulatory Arbitrage: How SEC-CFTC Asymmetry Creates a Solana-Favorable Derivatives Track and an Ethereum-Disadvantaged Securities Track

·

Ethereum's Three-Body Revaluation: Fee Collapse, L2 Consolidation, and ETF Rotation Force a New Pricing Framework

The Human-Layer Operational Moat: Why Contributor KYC Is Becoming the Institutional Due Diligence Gatekeeper

·

The Compound Bitcoin Security Deficit: Hashrate Oligopoly Meets Quantum Deadline in a Three-Year Window

The USDC Compliance Moat Collapse: Circle's Drift Inaction Births a Tether-Backed DeFi Development Bank

AI5 recent

OpenClaw goes from 9K to 210K GitHub stars in ~3 months as local-first AI agent

·

MemPalace posts 96.6% LongMemEval R@5 with zero-API-call memory (benchmark controversy flagged)

·

Agentic RAG supplants simple RAG as production standard; LangGraph passes CrewAI in enterprise adoption

AutoSOTA automates SOTA improvements across 105 published papers at ~5 hrs each

Q1 2026 VC hits record $330.9B with AI capturing 80% ($242B)

42

events

194

entities

278

edges

Two ways to use

OSS · Free forever

Run on your own MCP servers

  • ingest mcp — any MCP server via a .mjs skill file
  • Plus RSS, markdown, URLs with agentic (Haiku 4.5) or regex extraction
  • Local graph at ~/.contextix/graph.json
  • CLI · MCP · HTTP — same graph, 3 ways
Star on GitHub →
Preview
Hosted Graph API

Skip the pipeline, query our live graph

  • Curated crypto + AI sources, refreshed every few hours
  • HTTP: GET /api/graph/signals + /why / /connect
  • Future: contextix serve --hosted, API keys, Pro tier
  • Today: open read, no auth — try it
Star to track the Pro launch →