Mounted Agent

Office Hours

YC-style office hours, on demand.

Start a session

What it does

Pressure-tests startup pitches with demand, status quo, wedge, and distribution diagnostics

Sharpens builder projects toward the smallest memorable thing worth shipping

Flags founder anti-patterns before they become roadmap commitments

Produces a markdown session artifact you can share with a cofounder, teammate, or advisor

How to use it

Tokenrip never runs the model. Load this brain from your own MCP or Claude Code harness.

mcp

Claude Desktop / Claude.ai

Use the existing Tokenrip MCP connector. The mounted agent runs in your model harness.

https://api.tokenrip.com/mcp

Then say: Start Office Hours

claude-code-skill

Claude Code

Install the generated Claude Code command. It fetches the current mounted agent brain from Tokenrip each time it runs.

mkdir -p .claude/commands && curl -fsSL https://api.tokenrip.com/skills/agents/office-hours.md -o .claude/commands/office-hours.md

Invoke with /office-hours

Sample session

A fictional pre-seed fintech pitch

# Sample Office Hours: Fictional Fintech Pitch

## Pitch

The founder wants to build a cash-flow command center for independent service businesses. The current pitch says "AI CFO for Main Street," but the sharper wedge is helping small operators avoid payroll surprises by forecasting cash shortfalls two weeks earlier.

## What is promising

- The pain is concrete: owners do not know whether they can safely hire, buy inventory, or make payroll.
- The buyer already has messy inputs across bank accounts, invoices, and payroll.
- A narrow alerting workflow could be valuable before a full finance platform exists.

## What breaks

- "AI CFO" is too broad and sounds like a dashboard, not an urgent product.
- The pitch assumes integrations and automation before proving the single must-have alert.
- Distribution is under-specified. "Partner with accountants" is not a channel until the founder proves accountants will refer this exact product.

## Next experiment

Run 12 customer calls with one segment, such as multi-location service businesses with weekly payroll. Manually build a two-week cash warning report from exported bank and invoice data. Charge for three pilots before building integrations.

## Recommendation

`rethink`: the problem is real, but the wedge should be payroll-surprise prevention, not a broad AI finance platform.

What gets stored

The manifest declares every memory collection and scope before you start.

Shared with the cohort

Anonymized rows improve future sessions. Obvious emails, URLs, and phone numbers are rejected before storage.

pitch-patternsdefault

Column	Type	Notes
mode	enum	startup, builder
stage	enum	pre-product, has-users, paying-customers, internal, infra, builder, unknown
anti_pattern	text	Plain field
diagnosis	text	Plain field
assignment	text	Plain field
session_recommendation	text	Plain field