Get Started
Sign In Get Started

For AI agents · Free MCP server

Your agent is only as good as its context

Agents analyze whatever is in front of them. Before yours looks at an AI-exposed company, give it the current state of the AI cycle: what changed, who is arguing what, and the sources, all dated. The Luminix MCP server is free and needs no account.

Server URL https://www.useluminix.com/mcp

Works with Claude, Claude Code and other MCP clients. Setup takes a minute.

What your agent sees right now 
# AI cycle context, as of 2026-11-02

From the last 10 AI Cycle Monitor issues (2026-10-24 to 2026-11-02) and Luminix's published reports. Commentary and research, not investment advice.

## The read, and how it has moved

- 2026-11-02: S&P cut Meta's outlook to negative, forecasting 2027 capex above $225bn. Moody's calls AI debt stress contained; Gartner says $3.5tn of spend is at risk. Cash flows remain the question.
- 2026-11-01: Even a bull hedges: Ives calls Anthropic's IPO a 'perfect storm' as money tightens. Coatue steps back, Meta's CDS hits a record. TSMC still has pricing power.
- 2026-10-31: OpenAI cut its funding target to about $20bn at $1.25tn, the FT reports, as the selloff spread to chips. Bears got louder; Sacks reads Anthropic's S-1 as real productivity.
- 2026-10-30: Microsoft cut its capex guide to $170bn and Alphabet held, both under estimates; Reuters says Alphabet shed $500bn. Lenders get choosier: Ares avoids GPUs, Nvidia may backstop Nscale.
- 2026-10-29: Microsoft kept free cash flow positive; Alphabet fell 10% as $51bn of capex deepened its cash burn. Credit is the other pressure point: Bloomberg says AI debt yields are hitting double digits.

## Bulls vs bears

Over 10 issues: 24 bear stories, 13 bull stories.
- Making the bear case: Michael Burry, Ed Zitron, Apollo, Barclays, Ed Elson, Gartner
- Making the bull case: Dan Ives, Gavin Baker, All-In Podcast, Brad Gerstner, David Sacks, Ed Yardeni

… plus the latest by section, and the stories and reports for any company you name.

Live output of cycle_context. It updates each weekday at 7am ET.

Why context decides the answer

Today's models reason well. What they reason about is whatever lands in their context window, and they can't tell you what they never saw. Three gaps show up in almost every analysis.

1

The calendar gap

A model's knowledge stops at its training cutoff. In a story that moves every week, like AI spending, last quarter's picture leads to the wrong conclusion, and the agent states it with full confidence.

2

The fragment gap

Web search hands an agent whatever ranks: a snippet, an old article, a figure with no date. We see it in our own search data. In a sample of the Google searches where our reports appeared, 35% looked machine-written: quoted figures, long OR chains, lists of sites to skip. Agents are already hunting for context. Mostly they find fragments.

3

The one-side gap

Ask whether something is a bubble and search returns whichever side wrote the most pages. Unless the agent also sees the other side, attributed and dated, it repeats the first answer it found.

Same model, same question. Better context, better analysis.

Five tests for good context

Use them on any source you give an agent, ours included.

  1. Dated

    Every claim says when it was true. Without dates, an agent can't weigh old against new.

    Luminix: Every story, daily read and report carries its date, and cycle_context opens with an as-of line.

  2. Sourced

    Claims link to the original, so the agent can check them and you can audit the result.

    Luminix: Every story links to its source, and every report publishes the research threads behind it.

  3. Two-sided

    A live debate keeps both sides, attributed by name, instead of blending them into one average view.

    Luminix: Bears and bulls are tracked separately, with who said what.

  4. Maintained

    Someone keeps it current. A one-off scrape starts going stale the day it's made.

    Luminix: The AI Cycle Monitor is edited every weekday, and reports are refreshed when the story moves.

  5. Layered

    A short summary first, with depth on request, so the context window holds what matters.

    Luminix: cycle_context is the summary. Reports and their research threads are the depth.

Habits that improve any agent analysis

These work with any agent, on any topic, with or without Luminix.

  • Tell it today's date. Then ask it to flag anything older than a few months as possibly out of date.
  • Ask for a source and a date beside every claim. It's the fastest way to spot a guess.
  • Give it the counterargument before it concludes. Name the strongest opposing view, or point it at a source that covers both sides.
  • Separate what's reported from what's argued. A revenue figure and an opinion about it are different kinds of evidence.
  • Ask what would change its mind. A good analysis names the facts that would overturn it.
  • Load the summary first, then pull detail. Long, unfocused context dilutes the parts that matter.

Ways people use AI cycle context

Six jobs where your agent should know where the AI cycle stands before it starts. Copy a prompt and change the names.

Reading an earnings call

Analysts, journalists, operators

Before you summarize Microsoft's earnings call, get the AI cycle context from Luminix. Then tell me which open questions in the debate the call speaks to.

Vetting a vendor or partner

Procurement, engineering and strategy teams

We're choosing a GPU cloud provider. Get the AI cycle context for CoreWeave and list what the coverage says about its financing and customers, with dates.

Planning spend and contracts

Finance and infrastructure leads

We're planning next year's compute budget. Use Luminix to summarize what's happening to GPU supply, memory prices and data-center power, with sources.

Writing about AI

Writers, newsletter authors, students

Draft a 600-word explainer on whether AI spending is a bubble. Use Luminix for both sides and cite every claim with its date.

Weighing a job offer

Anyone joining an AI company

I have an offer from an AI lab. Pull what Luminix has on its revenue, burn and IPO plans, and list the questions I should ask before I decide.

Keeping a team briefed

Leaders who need the weekly picture

Every Monday at 8am, get the AI cycle context from Luminix and give me five bullets on what changed for Nvidia, Microsoft and Oracle.

Claude and ChatGPT can run a prompt on a schedule, so the weekly briefing can arrive without you asking.

What your agent gets

Six read-only tools. The server tells your agent to start with cycle_context and go deeper from there.

Everything it returns is also public on this site. The server hands it to your agent in a form it can use, with dates and links to cite.

cycle_context
Where the AI cycle stands: the latest read and how it has moved, bulls vs bears and who is making each case, and the latest on the labs, capex, chips and the money. Name a company or ticker to add the stories and reports that mention it.
search_reports
Luminix reports on a topic, each with its takeaway and a matching snippet.
get_report
A full report, with its research threads and how to cite it.
get_source_report
One research thread behind a report, with its citations.
latest_briefing
This morning's AI Cycle Monitor, as subscribers got it.
whats_changed
Everything since a date, filtered by section or by person if you like.

Connect in a minute

  1. Claude

    On the web or desktop, open Settings → Connectors → Add custom connector. Name it Luminix and paste the server URL.

  2. Claude Code

    Run this in your terminal:

    claude mcp add --transport http luminix https://www.useluminix.com/mcp

  3. Other agents

    Any MCP client that supports remote servers works. Add https://www.useluminix.com/mcp. There's no key and no account.

Then try: Use Luminix to get the AI cycle context, then tell me what changed this week.

Context, not calls

Luminix doesn't tell you or your agent what to buy or sell, and nothing is tailored to your situation. Everyone gets the same published research: what's happening, who is arguing what, and where it came from. The analysis your agent does with it is yours.

For informational purposes only. Not investment advice. Luminix is not a registered investment advisor.

Questions

What is an MCP server?

MCP (Model Context Protocol) is an open standard for connecting AI assistants to outside tools and data. Add an MCP server to Claude or another client and the assistant can call its tools when it needs them.

Does it cost anything?

No. The Luminix MCP server is free, and there is no account or API key.

What does it cover?

The AI cycle: the labs, hyperscaler capex, chips, the money behind data centers, and what the bulls and bears are saying, from the AI Cycle Monitor. It also covers every report Luminix has published on AI companies, markets and industries.

How current is it?

The AI Cycle Monitor is edited every weekday morning, and cycle_context updates with it. Every report shows when it was published and when it was last updated.

Why not just let my agent search the web?

It should, for anything Luminix doesn't cover. Luminix adds what search fragments miss: a maintained, dated view of the AI cycle, with both sides of the argument and links to the originals.

What do you see when my agent connects?

Only the requests your agent sends: the tool it called, what it asked for (such as a search phrase or a company name), the client name and a hashed IP address. Not your conversation. We delete these logs after 90 days. Read the privacy policy.

Is this investment advice?

No. Luminix publishes research and commentary. It doesn't recommend buying or selling anything, and Luminix is not a registered investment advisor.

Give your agent the context

Server URL https://www.useluminix.com/mcp