Setell-MCP · Coming summer 2026

Setell, inside
your favorite agent.

A vertical agent your agent can call.

Connect Setell to Claude Code, Claude.ai, ChatGPT, Cursor, or any MCP-aware agent. Drive your shop from the same tool you live in — same vertical brain, same per-relationship memory, same brand-matched output, three primitives any host can render.

Join the waitlistHow this fits with Claude SMB
v0 = stdio + Pro tier. HTTP + OAuth in Phase II.

What you'll be able to do

Three primitives. One vertical brain.

MCP servers expose three things, and clients render each differently. Setell uses all three for what they're designed for — not one overloaded tool surface.

Model-controlled
Tools

Actions your agent invokes.

Filter jobs, look up customers, pull revenue summaries, nudge a stale quote. Each tool is annotated with read-only or destructive hints so the host's approval UX is honest about what's about to happen.

“Boxx, find me Acme's last three quotes.” The agent calls setell_find_jobs and you get the list back inline.
Application-controlled
Resources

Context you attach.

Pickable URIs your host renders in its '@-mention' picker. Attach a job, a customer history, your brand settings, today's morning brief. The model reads them; you choose what it sees.

Paste setell://jobs/cm123 into Claude.ai and the full job state loads — versions, status, dealroom flags, the customer's relationship memory.
User-controlled
Prompts

Slash commands.

Discoverable templates that expand into structured prompts your host submits. The agent does the synthesis; the prompt scaffolds which tools and resources to compose.

/setell-triage-inbox walks your model through this morning's NEW jobs, pricing signals, and a recommended next-step per inbound.

Why this matters

A primitive Claude calls.
Not a thing Claude replaces.

The agent ecosystem isn't a single product. It's a stack of primitives that compose. Generalist agents handle the horizontal work — close the books, draft a campaign, review a contract. Vertical agents handle the work that doesn't generalize: per-vertical pricing memory, per-relationship customer memory, a deterministic revision engine, brand-matched outbound. That's Setell.

Setell-MCP is how that vertical brain shows up in your agent. Your bookkeeper drives it from Claude Code. Your operator runs the morning triage from Claude.ai. Your agent stack composes Setell with QuickBooks, Stripe, DocuSign, Gmail — whichever primitives it needs — without anyone opening the Setell UI. The Setell app remains the polished dashboard; the agent surface is where the workflow happens.

State converges across all three surfaces. A quote accepted from the app shows up in the next agent query. A follow-up sent from the agent appears in your Setell pipeline. Three doors. One canonical state.

Agent-first operators

Machine shop and contractor owners on Claude Max or Cursor who already live inside an agent all day. Setell stops being a separate app and becomes a tool their existing agent calls.

Intermediaries

Bookkeepers, fractional CFOs, and 'tech nephew' setups running quote-to-cash on behalf of multiple operator clients. One Claude Code workbench, many Setell tenants, switched by project context.

Hybrid users (trust-graduating)

Operators who start in the Setell UI and graduate routine actions to their agent over time — follow-ups first, status updates next, brand-matched drafts after that. The Setell app stays the dashboard; the agent becomes the workflow tool.

What ships at v0

A small, discoverable surface.

Seven tools, seven resources, five prompts. Read-mostly. Two low-stakes mutators — everything that touches outbound customer comms or financial truth stays UI/email-surface only until Phase II ships HTTP + OAuth.

Tools · 7
setell_find_jobs
filter by status, customer, date
setell_find_customers
search by name, email, company
setell_get_revenue_summary
period-bounded funnel stats
setell_list_templates
your quote templates
setell_get_pricing_signal
pgvector pricing memory
setell_send_quote_followup
low-stakes nudge — Trust-mode aware
setell_update_customer_memory
upsert per-relationship memory
Resources · 7
setell://jobs/{id}
full job state, version history, dealroom
setell://jobs/{id}/quotes/v{n}
specific quote version + evaluator verdict
setell://customers/{id}
customer profile + memory rows
setell://customers/{id}/history
every job, follow-up cadence, margin
setell://settings/brand
your voice, colors, terms
setell://insights/morning-brief
today's brief, computed on read
setell://health
connection + plan status
Prompts · 5
/setell-triage-inbox
summarize NEW jobs from the last 24h
/setell-stale-jobs
find SENT jobs past their cadence + draft follow-ups
/setell-weekly-revenue
trailing 7-day funnel + week-over-week
/setell-customer-history
load full history for a named customer
/setell-draft-followup
draft a follow-up — review before sending

v0 ships at the Pro tier, stdio transport, with the existing Setell extension key as auth. Send-quote, invoice creation, and the rest of the privileged surface graduate when Phase II ships HTTP + OAuth.

Get the install
when it ships.

Waitlist signups get first access, help shape the v0 surface, and lock in the Pro tier at the launch price.

Join the waitlistUse Setell in the app or Gmail today

Or use Setell in the app or Gmail today — the agent surface joins them, it doesn't replace them.

FAQ

What people ask first.

When does Setell-MCP launch?
Early access opens summer 2026. v0 is stdio transport with a curated read-mostly tool set plus two low-stakes mutators (follow-up nudges, customer-memory writes). HTTP + OAuth lands in Phase II, which is when send-quote, invoice creation, and the rest of the privileged surface graduate to the agent layer. Waitlist signups get first access and shape the v0 surface.
Will it work with Claude Code?
Yes — Claude Code is the primary launch target. The install is a single 'claude mcp add setell' invocation that drops the server into your config with your Setell extension key as an env var. Claude.ai desktop and ChatGPT desktop (newer versions with MCP support) follow the same pattern with a small JSON config snippet. Cursor and any other MCP-aware host inherit the same surface.
How is this different from a chatbot or a copilot in the app?
A chatbot lives inside one product. Setell-MCP lives inside your agent. You don't open Setell to use it — your agent calls Setell as a primitive while you're working somewhere else. Tools are model-invoked actions; resources are pickable context you '@' into the chat (like a job or a customer history); prompts are slash commands you trigger. Three primitives, one vertical brain, available wherever your agent runs.
Do I need a Pro plan?
Yes. Setell-MCP is a Pro-tier capability. The plan-gate fires at server boot — if your tenant isn't on Pro, the MCP server exits cleanly with an upgrade link. If you cancel mid-session, the next tool call returns a typed 402 with the same upgrade link. Free and Business plans get the Setell app + Gmail surface; the agent surface is Pro.
What about ChatGPT or OpenAI agents?
Setell-MCP is transport-agnostic by design. Anywhere MCP runs, Setell-MCP runs. ChatGPT desktop with MCP support works at v0. The Anthropic-specific bits are how we built the rest of Setell, not the agent surface itself. As more hosts adopt MCP we don't need a per-host fork — the same server answers all of them.
I'm not technical. Can I still use this?
v0 is honest about its audience — power users, agent-first operators, and intermediaries (bookkeepers, fractional CFOs, 'tech nephew' setups) who are already pasting MCP configs into Claude Code. If that's not you, the Setell app and Gmail surface stay primary. When Phase II ships HTTP + OAuth, the install collapses to one click in a browser — no config files, no Node prerequisite. Most non-technical operators will wait for Phase II. Both surfaces share the same vertical brain.
Want the full design memo? See how Setell + Claude SMB compose →