Work-Bench is a NYC-based $160M enterprise software Seed stage VC fund. If you are or know anyone thinking about founding an enterprise/vertical AI or infrastructure startup - we’d love to meet! Please reach out to chat.
💡 THE TIP
Your MCP has its own retention curve, and it looks nothing like your app’s.
Nobody has benchmarks for it yet. Apollo found theirs the hard way: after launching with Anthropic in March 2026, week 4 retention (user retention after 4 weeks) fell ten points and credits per team hit an all-time low. Why did this happen? A lack of product parity between their MCP server and web application. Their MCP exposed 20% of what the web app could do, so builders shipping real GTM automations hit a wall immediately.
Everything changed when they started to treat their MCP server as a real product. They shipped 30+ MCP features in twelve weeks, mapping much of what existed in the classic product over to the new MCP form factor. And it worked wonders, evidenced by story told by the numbers: week 4 retention +34%, new teams per week +47%, credits per team +135%, each a new all-time high, plus millions in ARR.
The lesson? While the API was a channel, an MCP is a true surface, and which demands a team to treat it like one.
Special shoutout to Tyler Phillips at Apollo for this tip!
✅ THE TO-DO
Open your MCP tool list and your product’s own navigation side by side. Write down every action a paying customer can take in the UI, then mark which ones an agent can actually reach. Sort the gaps by how often each gets used in the UI. That ranked list is your roadmap.
If you found this tip helpful, please share with a fellow b2b founder 😎
🤖 THE CLAUDE SKILL
Paste this into Claude with your feature list and your tool schema.
You're auditing the parity gap between our product UI and
our agent surface (MCP server or public API).
What a paying customer can do in our UI:
[PASTE YOUR FEATURE LIST OR NAV STRUCTURE]
What our MCP server / API currently exposes:
[PASTE YOUR TOOL LIST OR ENDPOINT LIST]
Do three things:
1. Build a parity table. One row per UI capability. Columns:
exposed to agents (yes / partial / no), and what's missing
if partial.
2. Rank the gaps by how central each capability is to the job
the customer hired us for, not by how easy it is to build.
Flag any gap that breaks a full workflow, since a dead-end
workflow churns harder than a missing feature.
3. For the top five gaps, write the sentence a churned user
would say out loud. Format: "I could do X in your app but
the agent couldn't, so I went back to the UI and stopped
using the integration."
Close with the single gap that, if fixed, unblocks the most
complete workflows.Do you have a GTM protip? Send our way here to feature in an future issue! If you try any of these tips and found them helpful, let us know - we’d love to hear from you.