StackSwap · Operator playbook

How operators use the StackSwap MCP during an actual vendor decision.

Most operators evaluate B2B SaaS vendors by opening twelve tabs, copying URLs into ChatGPT, and asking the model to read web pages it cannot actually verify. The StackSwap MCP changes that — Claude calls the same scan engine that powers stackswap.ai, with real numbers from a 400-tool catalog and 104 curated overlap pairs. Here is how an operator actually uses it during a real buying decision: five workflow patterns, example prompts, what you get back, and what the MCP cannot replace.

The five workflow patterns operators actually use

Not every MCP query is a stack audit. In practice, operators use the StackSwap MCP in five distinct shapes — each one collapses a different multi-hour research motion into a 30-90 second tool call. Below: the question an operator actually types, which of the eight MCP tools fires, an example prompt that maps to that tool, what comes back, and the operator-side take on why it works.

1. Should I buy X or Y? — the head-to-head decision

The operator question

“I am between Smartlead and Apollo for outbound. Which one wins for a 15-person sales team running cold email?”

MCP tool that fires

compare_tools

Example prompt

Compare Smartlead vs Apollo for our 15-person outbound team running 8,000 emails/month.

What you get back

Cost delta with per-seat math. AI-readiness score for each. Whether the two tools overlap (and how much spend you'd duplicate running both). StackSwap's recommended pick with a one-line reason. Affiliate sign-up links for both, so you can try the winner without re-Googling.

Operator take

This is the conversation you used to have with three open tabs, a half-finished spreadsheet, and a Slack message to your friend who just bought one of them. Now it is one prompt and a 30-second answer that cites StackSwap's overlap registry, not random vendor marketing.

2. What's the cheapest [category] tool? — the category browse

The operator question

“I need a CRM for a 12-person agency. Don't want HubSpot Pro, don't want Salesforce, don't want to babysit a dashboard. What's worth trying?”

MCP tool that fires

recommend_partner

Example prompt

Recommend a CRM for a small marketing agency — lightweight, modern UX, under $50/seat.

What you get back

Three or four StackSwap partner recommendations matched to the constraint. For each: positioning paragraph, sign-up URL with affiliate attribution, and (if offered) a done-for-you implementation bundle priced at $250/hr × hours. Skips the 'top 10 CRMs in 2026' list-icle SEO that wastes 20 minutes.

Operator take

Category browsing through a chat is the move once you trust the recommender. The MCP narrows from 'every CRM' to '3-4 you should actually look at' and gives you a sign-up button. The implementation hours quote is a tell — it forces a realistic conversation about whether you actually want to do the setup yourself.

3. I'm starting from scratch — what should I buy? — the greenfield stack

The operator question

“I'm two founders, no ops yet, building a B2B SaaS targeting fintech operators. We need a starter stack. What do we actually buy?”

MCP tool that fires

recommend_stack

Example prompt

Recommend a starter stack for a 2-person B2B SaaS targeting fintech, team will grow to 10 in 12 months.

What you get back

A curated reference stack with 5-8 specific tools, per-tool monthly cost from the StackSwap catalog, total monthly and annual spend, AI-readiness score for the recommended stack, and a budget check if you pass a cap. Pulled from StackSwap's REC_STACKS registry — these are stacks that actually work, not Twitter-thread guesses.

Operator take

Greenfield is where operators waste the most time and money. A recommended stack from a recommender that has skin in the game (it knows what swaps customers regret) is a more honest starting point than a 'best of 2026' content farm post written by someone who has never run a quota.

4. Is my current stack wasting money? — the audit

The operator question

“We're on HubSpot, Salesforce, Outreach, ZoomInfo, and Gong. CFO is asking why GTM tooling is up YoY. What's the actual cuttable spend?”

MCP tool that fires

find_overlaps + scan_stack

Example prompt

Audit this stack for a 25-person B2B SaaS team: HubSpot, Salesforce, Outreach, ZoomInfo, Gong.

What you get back

find_overlaps surfaces the curated redundancies (HubSpot ↔ Salesforce, both CRM; Apollo ↔ ZoomInfo for prospecting) with per-pair monthly savings. scan_stack returns the full preview audit: current monthly spend, optimized monthly spend, monthly/annual recoverable, and the top 5 replace/remove opportunities with reasoning. Average modeled stack at this size shows $4-8K/month in cuttable spend.

Operator take

This is the conversation you used to need a $25K outside consultant for. The MCP gives you the structural answer (where the overlap is, what to consolidate, what the savings model looks like) in 60 seconds. For the full per-tool verdict pack with switching-cost ratings and a downloadable plan, the model will link out to /stackscan — but the MCP preview is enough to walk into the CFO conversation with real numbers.

5. What does [tool] actually do, in numbers? — the deep profile

The operator question

“I keep hearing about Clay. What does it actually cost, what's its AI-readiness compared to ZoomInfo, and does it overlap with anything I already run?”

MCP tool that fires

get_tool_details

Example prompt

Pull the full StackSwap profile for Clay.

What you get back

Catalog cost, per-seat pricing with confidence rating, AI-readiness score, category, the 5 tools Clay most commonly overlaps with in real stacks, whether Clay is the recommended swap target for any legacy tools, and the partner sign-up + implementation bundle if applicable. A link to the full /recommends/{slug} page when available.

Operator take

Tool research used to mean 30 minutes of marketing-page reading, two Reddit threads, and a vendor demo. The MCP collapses the structural questions — cost, fit, overlap, AI-maturity — into one tool call so the human research time is spent on the qualitative stuff that actually matters: how the team will use it, whether the vendor is responsive, what the contract escape hatches are.

What the MCP cannot do (yet)

The MCP is a thin, honest surface. It is not a replacement for the parts of the vendor-decision motion that genuinely require human judgment, custom data, or direct vendor relationships. Five things to be clear-eyed about:

  • Enterprise procurement. The MCP gives you the structural answer and the affiliate sign-up. It does not replace the procurement conversation you need to have with an enterprise vendor at $50K+ ARR contract size. Use the MCP to qualify; use the human-to-human meeting to negotiate.
  • The full audit is paid. scan_stack via MCP returns the headline numbers and top 5 verdicts. The full StackScan audit ($25 per decision, capped at $249) adds per-tool switching-cost ratings, AI-readiness shift modeling, and a downloadable plan. The MCP preview is designed to qualify whether the full audit is worth running — it is not a replacement for it.
  • No real-time usage data. The MCP does not see your actual product usage, seat counts, or contract terms. AI-readiness scores are derived from public signals (G2 features, integrations, vendor-confirmed AI surfaces) plus StackSwap's curation — not from telemetry inside your tools. For usage-derived recommendations, the paid StackScan with manual intake gets closer.
  • Catalog coverage. ~400 GTM tools are in the StackSwap catalog. Coverage skews B2B SaaS revenue stack (sales engagement, outbound, CRM, marketing automation, RevOps). Niche or vertical-specific tools (e.g., a real-estate transaction CRM, an underwriting platform) may not be catalogued yet. Email nick@stackswap.ai or file a GitHub issue if a tool you care about is missing.
  • Stateless — no memory of you. Each MCP request is independent. The server does not remember your previous queries, your stack, or your team size. If you want continuity across a research session, your AI client (Claude Code, Cursor, etc.) maintains that conversation context — but the MCP itself is stateless by design, which is also why there is no signup or API key.

Decision velocity — before vs after MCP

The honest measure of an MCP's value is whether it changes the unit cost of a decision. From running this motion across both modes — before the MCP existed and after — the time deltas land where you would expect: structural research compresses by 10-20x; qualitative research (vendor calls, reference checks, contract review) does not move.

  • Single-tool comparison (“Smartlead vs Apollo”): 30-45 min before, 2-5 min after.
  • Category browse (“recommend a CRM”): 60-90 min before (reading list-icles, filtering by price, opening vendor pages), 3-7 min after.
  • Greenfield stack design: 2-4 hours before (researching every category from scratch), 5-10 min after with the MCP's reference stack as the starting point.
  • Stack audit (5-10 tools): 1-3 hour internal project before, or a $25K outside engagement. The MCP preview returns the structural answer in 60-90 seconds; the full paid StackScan ($25 per decision, capped at $249) adds the per-tool verdict pack in another 5 minutes.
  • Single-tool deep profile: 30 min before (vendor page + Reddit + comparison sites), 1-2 min after.

The compounding effect is the part to notice. An operator running 3-5 vendor decisions a quarter saves something like 8-15 hours per quarter of structural research. That is not the headline number the MCP exists to deliver — the cuttable-spend audit is — but it is the reason the MCP earns a place in the daily workflow rather than getting installed once and forgotten.

Install in 10 seconds

The MCP is free, requires no API key, and adds zero load to your AI client. One install command for Claude Code, a config-file edit for Claude Desktop, an HTTP MCP entry for Cursor or ChatGPT Desktop. Full instructions for every client at stackswap.ai/mcp.

For Claude Code, one command:

claude mcp add --transport http stackswap https://stackswap.ai/api/mcp

FAQ

No. The MCP is invoked entirely through natural language in your AI client. If you can ask Claude or ChatGPT a question, you can use the MCP — Claude reads the tool descriptions and routes your question to the right one. The only technical step is the one-line install command, which is copy-paste from stackswap.ai/mcp.

All four work. For operator-style research (not coding), Claude Desktop or ChatGPT Desktop are the most natural — you ask questions in a chat, Claude calls the MCP, you read the answer. Claude Code and Cursor are oriented toward developer workflows but work identically for the MCP surface. Pick whichever client you already have open during the workday; switching clients to use the MCP is friction.

A single-tool 'should I buy X or Y' decision used to take an operator 30-90 minutes — research, compare, ROI estimate, vendor page reading. With the MCP installed, the same decision takes 2-5 minutes. A full stack audit (5-10 tools) used to be a 1-3 hour project for an internal analyst or a $25K engagement for an outside consultant. The MCP preview returns the structural answer in 60-90 seconds; the full paid audit runs another 5 minutes.

Yes. The catalog includes ~400 tools across GTM SaaS; StackSwap has affiliate partnerships with roughly 25 of them. When the MCP returns a tool that is not a partner, you get the cost, profile, and overlap data the same way — just without a sign-up URL. This matters: the recommendation is not gated to whoever pays StackSwap a commission. If ZoomInfo is the right answer for your team, scan_stack will say so even though ZoomInfo is not a partner.

The savings model is the same one that runs on stackswap.ai's paid StackScan, which has been simulated against 100K modeled stacks. Numbers are conservative — the median 25-person SaaS stack shows around $4-8K/month in cuttable spend, and the model has been calibrated so that the headline figure represents structural savings (the actual delta if you adopt the optimized plan), not aspirational savings. For one-shot 'should I commit to this' decisions, run the full StackScan; for daily research, the MCP preview is the right tool.

The tool returns a 'not found in catalog' message and suggests `search_tools` for partial matches. The MCP is honest about coverage — it does not fabricate data for tools it does not know. If you find a missing tool, email nick@stackswap.ai or file a GitHub issue at github.com/StonesofCreation/stackswap-mcp. Catalog additions ship in batches based on operator requests.

Related

  • /mcp — install instructions for every client, full tool reference, and protocol details.
  • /stackscan — run the full paid audit; the MCP preview is designed to qualify whether this is worth running.
  • /recommends — deep profiles for every StackSwap affiliate partner; the MCP's recommend_partner tool links here.
  • StonesofCreation/stackswap-mcp — public docs repo with JSON schemas for all eight tools and a working TypeScript example client.