Ask the whole channel.
Get the evidence.
I took all 124 videos of The Brülosophy Show, one of the best home-brewing channels on YouTube, and turned everything said in them into a knowledge graph: every claim, every number, every experiment, tagged with who said it, in which video, at what second. Then I put a public MCP server in front of it. Add one URL to Claude or ChatGPT and your assistant answers brewing questions from what was actually said on the channel, with a timestamped link on every answer so you can go watch the moment yourself.
It is free, it needs no account, and it is not a product. I built it because I wanted to ask a channel I like a question and get the real answer back. Now anyone can.
A brewing channel your
assistant can actually use.
Ask ChatGPT or Claude a brewing question today and you get a confident average of the internet. Ask it with this server connected and you get what Martin Keen said about it on The Brülosophy Show, at minute eleven of a specific video, with the numbers he measured and the result of the blind tasting that settled it.
That is the whole idea. Brülosophy is known for its exBEERiments: split a batch, change one variable, and let blind tasters try to pick the odd beer out. The answers are careful, honest and often surprising, and the most common finding is "no difference". This graph keeps that honesty intact. When the tasters could not tell two beers apart, the server says so, instead of implying an effect that was never found.
Not a product.
A gift to the hobby.
This server was not built for commercial purposes. There is no plan, no trial, no email gate and no login. You add the URL, you ask, you get answers. If you stop using it, nothing happens. That is the only commitment: none.
All the content belongs to Brülosophy and the people who appear on the channel. The graph is an index. Every claim carries a link back to the exact moment in the original video, so the server sends you to the source rather than replacing it. If you find it useful, go watch the videos and subscribe to the channel. That is the best way to say thanks.
Everything said on the channel,
made queryable.
A claim is one thing a person said that can be true or false: "a slow cold crash gives better head retention". Each one is stored with who said it, the video, the second it starts, whether it is stated as working, failing or making no difference, and the scope it applies to, so advice about an IPA is never quietly served for a Pilsner question. A measurement is a number said out loud (a temperature, a time, a gravity, a pH) kept in its original unit and converted, so "68 Fahrenheit" and "20 Celsius" answer the same question. An exBEERiment is a full record of a blind test: the variable, the two beers, how many tasters, how many picked the odd one out, and whether that counts as a real difference.
The questions a brewer
actually has.
Anything you would type into the channel's search bar and hope for the best, you can now ask in plain words. A few of the kinds of questions it is good at:
- Does purging the keg with CO2 change the beer?
- Can tasters tell a 44% smoked malt Rauchbier from a 63% one?
- Is homemade invert sugar different from plain sugar in a Tripel?
- What temperature for a Festbier primary?
- What mash pH did they hit on the oatmeal stout?
- What OG and FG did the Belgian Tripel finish at?
- My lager tastes of green apple. What causes that and what fixes it?
- Why is my hazy IPA turning brown?
- Is this plan safe, or is there a warning I should hear first?
- Which steps did the blind tasters never notice?
- Does a 30-minute boil hold up against a 60-minute one?
- What does the Short & Shoddy series say I can drop?
- Give me the grain bill and hops for their Victory malt Festbier.
- What is Weyermann smoked malt like, and what substitutes for it?
- Compare pickling lime against untreated mash water.
- What has John Palmer said about mash pH on the show?
- Where do Jordan Folks and Will Lovell disagree?
- How has the channel's view on cold crashing changed over time?
Here is what an answer looks like once the server is connected. Every number below is real, and every link opens the moment it was said:
Does purging the keg with CO2 actually make a difference?
It depends which purge you mean, and the channel tested three. Transferring a hazy IPA into a fully purged keg (filled with sanitizer, pushed out with CO2) versus a non-purged one: 22 blind tasters, 14 picked the odd beer out where 13 were needed. A real difference, and 9 of the 14 preferred the purged keg. significant
Purging only the headspace after filling, 15 times, on an American stout: no noticeable difference, and no signs of oxidation in either beer. no difference
Purging an empty keg with CO2 versus the sanitizer-fill method, on a German Pilsner, tested at one month and again at three months: tasters could not reliably tell them apart either time (10 of 22 at three months, 12 needed). no difference
[1] Does PURGING THE KEG Have A Perceptible Impact On Beer? · 4:05 [2] same video · 7:39 [3] same video · 11:43
Ask for the sources and the server draws them: the ingredients, styles and steps involved, how they relate, the claims underneath, and the videos they came from, as a card inside the chat.
Start with brew_answer.
Reach for the rest when you know what you want.
An MCP server exposes tools, and your assistant picks the right one for the question. You never call them by name, but knowing what is there tells you what kind of question the graph answers well. Every tool returns claims, not prose: polarity (works, fails, no difference), evidence type, the numbers stated, the experiment statistics, corroboration across videos and a timestamped YouTube link.
| Tool | Use it for | What it does |
|---|---|---|
| start herebrew_answer | Answer a brewing question in one call | Combines claim search, exact measurements, exBEERiment verdicts, fault diagnosis, no-difference findings and safety warnings as the question needs, and draws the sources card itself. |
| searchask_brewing | Ask the knowledge graph | Hybrid semantic search, reranked, expanded through the graph. Pass a style and advice scoped to another style is flagged, never served silently. |
| numbersmeasurements_for | Measurements for a style or process | "What temperature for an IPA primary?" as one query over 4,100 unit-normalised measurements, scope expanded through the style hierarchy. |
| evidenceexperiment_results | exBEERiment results | The split-batch triangle tests: tasters, correct picks, p-value, significance, and the verdict as stated. |
| troubleshootdiagnose_fault | Diagnose an off-flavour or fault | Causes, what prevents it and what fixes it after the fact, each backed by the claims that say so. |
| reciperecipe | Recipe from the graph | Grain bill, hop schedule, yeast and the target numbers for a recipe the channel brewed. |
| ingredientingredient_profile | Ingredient profile | A hop, malt, yeast or adjunct: what it does, what it is typical for, what it substitutes, and where it was used. |
| a vs bcompare | Compare two things the channel tested | Two ingredients, techniques or pieces of gear side by side, with the tests that put them head to head. |
| shortcutswhat_can_i_skip | What can I skip? | The channel's signature result: the steps and variables blind tasters could not tell apart. |
| attributionwho_said_what | Claims by speaker | What Martin Keen, Will Lovell, Jordan Folks, John Palmer or any guest actually said, with timestamps. |
| timelinehow_views_changed | How views on a topic changed | Claims on a topic in date order, including where a later video disagrees with an earlier one. |
| safetysafety_check | Safety check for a brewing plan | Pressure vessels, glass, sanitiser contact, CO2 in enclosed spaces: the safety-critical claims that touch your plan, surfaced regardless of rank. |
| sourcesshow_answer_sources | Show where an answer came from | Renders the entities, relations, claims and videos behind an answer as an in-chat card. Called once, after the specialised tools. |
Prompts that chain the tools for the common jobs.
| Prompt | Title | What it does |
|---|---|---|
| ask | Ask the Brülosophy graph | A question, answered from the evidence with citations and the sources card. |
| plan_brew_day | Plan a brew day | Recipe, target numbers, shortcuts that are safe to take, and safety notes for a style. |
| troubleshoot | Troubleshoot an off-flavour | A symptom in, causes, prevention and fixes out, with evidence. |
| skip_check | Can I skip this step? | What the channel's blind tests say about dropping a step. |
| brulosophy://glossary | Resource | The style hierarchy, stage vocabulary, measurement kinds, series formats and speakers: the terms the tools understand. |
Nothing is generated
on the server.
This matters more than any feature. The server never writes an answer. It finds the claims that match your question, expands through the graph to pull in the numbers, the experiments and the related claims, and hands that evidence to your assistant with the links attached. Your assistant writes the reply from that evidence and shows where each part came from.
Find
Your question is matched against every claim on the channel, by meaning and by words, then the best matches are ranked.
Expand
Each matching claim brings its neighbours: the measurements stated, the experiment it belongs to, claims that agree or contradict it from other videos.
Cite
Everything goes back with the speaker, the video and the timestamp. Your assistant answers from it, and you can click through to check.
Two honesty rules are built in. If the tasters could not tell the beers apart, the answer says "no difference" rather than dressing it up. And if a claim was made about one style, it is marked out of scope when you ask about another, so it is shown as context, never as advice.
Two minutes
to add it.
The server speaks the standard MCP protocol, so it plugs into any assistant that accepts a custom connector. No key, no login, no account with me. This is the address:
https://brulosophymcp.3nuggets.io/mcpPaste it into any MCP client. Authentication: none. Read-only: the server only ever returns claims from the graph.
Claude
- Open Settings → Connectors and choose Add custom connector.
- Paste
https://brulosophymcp.3nuggets.io/mcp. No login is needed. - Ask: "Does cold crashing slowly give better head retention?"
ChatGPT
- Settings → Apps & Connectors → Advanced → enable Developer mode.
- Create → MCP server URL
https://brulosophymcp.3nuggets.io/mcp, authentication None. - Ask the same question and check the link it cites.
Any other MCP client: Streamable HTTP, stateless, no session id required.
How answers are built. Nothing is generated on this server. Each tool returns the claims that match, and your assistant writes the answer from that evidence and shows where it came from. Brülosophy's most common finding is no difference, and the tools report it as such rather than implying an effect.
Transcribe. Read. Structure.
Then check everything.
The build is the same method I use for every digital brain, pointed at a YouTube channel instead of a company. The short version:
Collect the words
Every video's title, description and timestamped transcript was pulled through an API. No video was downloaded, and the transcripts are the channel's own captions.
Read every video with AI
Claude read each transcript, split it into topics, and pulled out every claim, number, ingredient, step and experiment, with who said it and at what second. Every result had to pass a strict contract before it was allowed into the graph: numbers only where the transcript says them, timestamps only inside the right segment.
Make the numbers comparable
Every measurement was kept as spoken and converted to a standard unit, so questions about temperature, time or gravity work across the whole channel regardless of how the number was said.
Audit the whole graph
With all 124 videos in, the graph was checked end to end against the transcripts. Duplicate people were merged (the host had turned up under three spellings from the captions), sponsor reads were separated from real claims, and a few converted numbers were corrected. Then a set of test questions was run against it until every one returned the right video.
Then the MCP server went in front of it, read-only, on a plain URL anyone can add. The whole thing is my usual stack, which is the point: the same tools that build a brewing brain for a hobby build a business brain for a company.
Before you add it.
Is it free?
Yes. Public, free, no signup, no email, no login. It was built as a gift to the home-brewing community, not as a commercial product, and there is nothing to commit to. Add it, use it, drop it whenever you like.
Does it make things up?
No. Nothing is generated on the server. It returns the claims that match your question with the speaker, the video, the timestamp and any numbers or experiment results. Your assistant writes the answer from that evidence and shows the links, so you can check every part of it against the original video.
What if the channel found no difference?
Then that is the answer. Brülosophy's most common finding is "no difference", and the graph reports it that way. If the tasters could not tell two beers apart, you are told so, with the numbers, instead of being handed advice that was never supported.
Which assistants does it work with?
Claude, ChatGPT and Claude Code today, and any other assistant that accepts a custom MCP connector over a URL. The install steps are above.
Who owns the content?
Brülosophy and its contributors. The graph is an index with links back to the original videos. It sends people to the channel, it does not replace it. If it earns you a better beer, go watch the videos and subscribe.
Can you build one of these for my business?
Yes, and that is the real reason this exists. The same method that turned a brewing channel into a queryable brain turns your documents, calls, recordings and know-how into one, and the same kind of MCP server puts it inside the Claude or ChatGPT your team already uses. We build it together in a Sprint, and you keep it.
Your business has
a channel like this too.
Hundreds of calls, documents, recordings and answers your team already gave once. I turn that into a brain your AI can query, and put an MCP server you own in front of it, so ChatGPT or Claude answers from what your business actually knows, with the source attached. Built live with you, in a Sprint.