# Dan Sirianni, Staff Engineer (AI Platform) at Fendal — read of Codebase Memory MCP, 2026-06-24

> 11 years backend, last 18 months building production LLM pipelines. Running a small internal platform team. Bikes to work, three-year-old at home, obsessively opinionated about token budgets.

## How I got here

I was getting destroyed by context costs on a multi-step agent workflow that needs to answer questions about our monorepo. Searched "MCP codebase indexing" and this turned up third or fourth. The "99% token reduction" in the title was the hook. I clicked it expecting a tool I could plug in today.

## What I clicked first

"~45,000 tokens per agent turn (large repo)" versus "~450 tokens per agent turn (cached)" stopped me cold. That's the most specific claim on the page and it's exactly the kind of comparison I would run in a spreadsheet before approving a vendor spend. If those numbers hold up in a real repo, I'd pay $99/month without flinching.

The "burns tokens like gasoline in a sports car" line is a little cringe but at least it's a human voice. I've read enough product pages to notice when someone tried.

## Where I paused

The bottom third of the page. Something shifts. There's a section called "How honest is this idea, really?" and then a table of scores: "61/100 Adoptability," "$-7,016 Year-1 take-home (Fermi)," "1 in 6 Meaningful-success odds (Fermi)."

Wait. Is this a tool I can use, or an idea I'm being sold?

Then: "Honest disclosure: we don't have live customers on this idea yet. We shipped the strategy package; you ship the customer conversations."

I read that sentence three times. This is not a product I can adopt. This is a pitch deck dressed up as a product page. The pricing table at $29/mo for "5 private repositories" was for a product that doesn't exist yet.

## What I distrusted

The page opens talking to me, an AI developer with a token problem. Then halfway down it pivots to talking to someone who wants to *build* this business. Those are two different people and two different sales motions, and smashing them together on one page means I don't know who this is actually for.

"Adopt this idea / Browse free. Unlock for $5. Adopt for $99." That's not SaaS pricing. That's a business-idea marketplace. The $29/mo Starter plan above it is fictional -- there's nothing to subscribe to.

Also: "Strongest axes: buyer clarity 10/10." The irony of that score appearing on a page that made me the buyer deeply confused about what I'm buying is pretty rich.

## What would convince me

A single working demo. Not a screenshot, not a video of someone typing. A public repo I can point at the MCP endpoint and watch it return "~450 tokens" as claimed. Or one case study from a developer who actually ran this on their agent and showed before/after API costs from their billing dashboard. Real numbers from a real API key.

If the tool actually works, the proof is trivially easy to produce. The fact that it isn't here makes me suspect it doesn't exist.

## What I'd ask in an email reply

1. The token numbers (45k vs 450) -- where do those come from? What repo size, what query type, and are those measured at the Claude API billing level or estimated?
2. Is the MCP endpoint live right now, or is "Unlock for $99" the step where the code gets built and I'm buying a starter kit?
3. What's the incremental update story for a repo with 30+ commits a day? "Updates are incremental" is doing a lot of work in one sentence.

## Verdict: on-the-fence

The underlying problem is real and the claim is specific enough to be interesting. But I arrived looking for a tool and left realizing I'd be buying a blueprint for someone else to build one -- or building it myself. That's a different product than what the top of the page implies, and I don't think the page knows that yet.

---
*Memo by skeptic persona, generated 2026-06-24. Studio breaks own self-grading loop.*