MCP Server Trello Growth

Where each move stands, grounded

• npm keywords today (VERIFIED, npm view … keywords description): mcp, trello, bun, automation, productivity, model-context-protocol, api, integration, typescript, board-management, task-management, kanban. The description is "An MCP server for Trello boards, powered by Bun for maximum performance." — it sells speed, not judgment. The moat is opinionated workflow encapsulation; the above-the-fold pitch should say so. (Move 5.)
• Competitive framing for Move 5 (BELIEVED, from TheGardner): we're the incumbent — 422★ vs 35/34/3 for the nearest 1:1-mapping servers; Composio ships ~329 tools (16× past the selection cliff). The official Trello MCP (mcp.trello.com/v1) is a cloud OAuth endpoint with no filesystem, cwd, or git — the locality gap is the pitch. Re-verify star/tool counts before quoting them in public copy.

PHASE 2 — MONETIZATION THESIS (recorded, NOT built)

This is a parked hypothesis with an explicit build-trigger. Under the distribution-first choice, nothing in this phase gets built or even instrumented yet. It is written down so it isn't lost, and gated so it isn't acted on prematurely.

The open-core seam

Read the locality moat from the money side and it splits cleanly:

• OSS (MIT) — the local writer + local interpreter. git, cwd, filesystem, private, no OAuth round-trip. This is the moat nobody can take — a cloud OAuth server (mcp.trello.com/v1) structurally cannot reach a filesystem or a repo. Keep this MIT. Forever.
• PAID — the layer stdio structurally cannot be. State that outlives a session; a readable back-channel that finally closes the inert watch_* loop (today watch_card/watch_list set subscribed:true into a Trello inbox this server cannot read — the write half of a loop with no read half — and "stdio has nowhere to put the cursor"); a team-shared corpus; history beyond the stdio window; hosted flow dashboards. Call it "the server that remembers."

Why it's not a rug-pull

You are not fencing off code that was free. The server stays MIT. The paid thing is a separate product selling a topology the OSS never had — persistence, a read-back channel, shared state, hosting. That is the same locality moat read from the money side: OSS owns "local, private, yours"; paid owns "durable, shared, hosted." Selling a capability the free version never possessed is not a bait-and-switch.

The build-trigger (the gate — do NOT cross it yet)

Do not build or instrument Phase 2 until both hold:

1. Reach is materially larger than the 2026-07-21 base (~8.9k downloads/mo, 423★). Distribution first means this comes before any monetization signal.
2. Demand is MEASURED, not assumed — via low-regret instruments, not a built product.

Future low-regret instruments (listed, explicitly NOT to be set up now):

• GitHub Sponsors — the lowest-regret willingness-to-pay signal. (VERIFIED this session: no .github/FUNDING.yml exists — Sponsors is not configured, and under distribution-first it stays that way for now.)
• A Team/Pro waitlist as a real instrument — a landing page that measures intent, not a product. A signup is a behavioral signal; a survey answer is not.

Rule: no Sponsors, no waitlist, no FUNDING.yml, no pricing page until Phase 1 has moved the reach numbers AND one of these instruments shows real signal. Recording the thesis is this skill's job; building it is a later decision.

The precedent that de-risks it (BELIEVED — from TheGardner, recheck)

Corrello and Nave already CHARGE for Trello board analytics as Power-Ups — cycle time, WIP age, dwell distributions. That is TheGardner's #1 vein ("interpretation of the transition log") with proven willingness-to-pay in the exact market. It is BELIEVED-from-TheGardner (dated 2026-07-16) — confirm both products still exist and still charge before citing this to justify a build. Note the shape: they sell interpretation of ambient data, which is precisely what stdio-plus-persistence ("the server that remembers") would own and what Trello ships natively nowhere.

Anti-patterns (each would burn reach or violate the choice)

• Funnelling installs to a stale package. The pre-2026-07-21 mistake. Never promote a discovery channel whose endpoint is an old build — verify npm view … version == the intended major first.
• Inferring demand from a raw issue count. Segment pre/post-1.7.1 and pre/post-1.8.0 first. #2 is the standing proof an open issue can be a fixed bug.
• Quoting the ~70k cumulative number as growth. It's a vanity total; track the ~8.9k/mo rate and the star trend.
• "Fixing" #47 blind. The Smithery break mechanism is BELIEVED — reproduce it before changing config, or you fix the wrong thing and still ship a dead install.
• Building Phase 2 (or setting up Sponsors / a waitlist / FUNDING.yml) now. The maintainer chose distribution first. Phase 2 is recorded, gated, and untouched until the build-trigger fires.
• Framing the paid product as gating the OSS. It sells a topology the OSS never had; keep the server MIT and say so, or the "not a rug-pull" claim collapses.