StackSwap · Model Context Protocol

Eight StackSwap tools, one URL, inside any MCP-compatible AI client.

Search ~400 GTM tools, find curated overlaps in a stack, suggest AI-native swaps, compare vendors head-to-head, run preview spend audits, pull greenfield stack recommendations, and get partner sign-up links with implementation hours — without leaving Claude Code, Claude Desktop, ChatGPT Desktop, or Cursor. Free, no API key, stateless JSON-RPC over HTTP.

Install

Claude Code (CLI)

One command in any terminal:

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

Claude Desktop

Add this block to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows), then restart Claude Desktop:

{ "mcpServers": { "stackswap": { "url": "https://stackswap.ai/api/mcp", "transport": "http" } } }

Cursor / ChatGPT Desktop / other MCP clients

Add a new HTTP MCP server with URL https://stackswap.ai/api/mcp. No authentication. The endpoint responds to GET with a server descriptor for sanity-checking and to POST with JSON-RPC requests. A working TypeScript example client lives in the public docs repo.

Tools exposed

  • search_tools — Fuzzy-search the StackSwap catalog of ~400 GTM tools by name. Returns each match with its monthly cost and (when applicable) a partner sign-up link.
  • find_overlaps — Pass a list of tools in a stack. Returns the curated overlap pairs (104 hand-verified) with estimated monthly/annual savings if one tool is consolidated.
  • suggest_swaps — For each tool supplied, returns the AI-native replacement StackSwap recommends — with annual savings and reasoning. Skews toward legacy → modern (Outreach → Smartlead, ZoomInfo → Apollo, etc.). (Smartlead deep profile →)
  • scan_stack — Runs a preview StackScan. Returns current monthly spend, monthly/annual recoverable, and the top 5 replace/remove opportunities. Includes a link to the full paid audit.
  • recommend_partner — Given a need (e.g. 'outbound', 'CRM', 'automation'), returns the recommended affiliate partner with sign-up URL, positioning, and (when offered) a done-for-you implementation bundle at $250/hr. (Apollo deep profile →)
  • compare_tools — Head-to-head comparison of two named tools. Returns cost delta, AI-readiness scores, overlap status, swap-registry signal, and StackSwap's recommended pick with reasoning. Use for 'X vs Y' decisions.
  • recommend_stack — StackSwap's reference starter stack for a given industry vertical. Returns a curated tool list with per-tool cost, total monthly/annual spend, AI-readiness score, and partner sign-up links. Use for greenfield 'what stack should I buy?' queries.
  • get_tool_details — Full StackSwap profile for a single tool: catalog cost + per-seat pricing with confidence, AI-readiness score, category, common overlaps, swap-registry status, partner link + done-for-you bundle if applicable. (Instantly deep profile →)

What this is built for

GTM engineers and operators who work inside Claude or ChatGPT and want StackSwap's tool data + scan compute available as native tool calls during a conversation. Ask “what overlaps if I'm on HubSpot, Salesforce, and Outreach?” or “what cold-email tool should I try?” and the model can answer with real StackSwap data instead of hallucinating.

The MCP is intentionally a thin surface. For the full audit (per-tool verdicts, switching costs, AI-readiness shift, downloadable plan) the model will link out to /stackscan. For done-with-you implementation work, it links to /recommends.

FAQ

The StackSwap MCP is a Model Context Protocol server that exposes eight read-only tools from stackswap.ai — tool search, overlap detection, AI-native swap suggestions, a preview StackScan, partner recommendations, head-to-head tool comparisons, greenfield stack recommendations, and single-tool deep profiles. Once installed, MCP-compatible clients like Claude Code, Claude Desktop, and ChatGPT Desktop can call these tools directly during a conversation, so the user can ask 'what overlaps in my stack?' or 'should I pick Salesforce or HubSpot?' without leaving the chat.

Free. The MCP is a distribution layer for stackswap.ai — the value capture happens downstream when users either click an affiliate partner link, run the full paid StackScan ($25 per decision, capped at $249), or hire StackSwap for a done-for-you implementation bundle. There is no API key and no rate limit beyond reasonable use.

No. The free tiers of Claude Code, Claude Desktop, Cursor, and ChatGPT Desktop all support MCP connections. The StackSwap MCP server itself requires no API key, account, or sign-up. Total out-of-pocket cost to install and use: zero dollars. You only pay if you decide to run the full paid StackScan or buy a done-for-you implementation bundle downstream.

Each tool call sends the function name and the arguments you (or the LLM on your behalf) pass — e.g., a list of tool names for the find_overlaps query. The server does not store conversation data or user identifiers; it's stateless JSON-RPC. The server does see your IP address at the HTTP level, standard for any web request, and that's not persisted beyond standard server logs. Affiliate links surfaced in tool responses carry utm_source=mcp so partner clicks are attributable, but no identifying data flows back to StackSwap.

The MCP scan_stack tool returns the monthly/annual recoverable spend and the top 5 replace/remove opportunities — enough to see whether your stack has meaningful overlap. The full paid StackScan ($25 per decision, capped at $249) adds per-tool verdicts with switching cost and confidence ratings, AI-readiness shift, a downloadable plan, and the canonical Recommendation Council. Use the MCP preview to qualify the opportunity; use the paid scan when you are committing to action.

Affiliate links returned by the MCP include UTM tags (utm_source=mcp, utm_medium=tool_response, utm_campaign={tool_slug}) so partner clicks are attributed back to the MCP surface separately from on-site clicks. The endpoint itself doesn't store conversation data or user identifiers — it's stateless JSON-RPC.

Yes. The endpoint speaks standard JSON-RPC 2.0 over HTTP. A GET request to https://stackswap.ai/api/mcp returns the server descriptor (name, version, tool list). A POST with a JSON-RPC envelope invokes any of the eight tools. The public docs repo at github.com/StonesofCreation/stackswap-mcp includes a ~50-line TypeScript example client that demonstrates the protocol end-to-end and works against the live endpoint with no API key.

The server speaks MCP protocol version 2025-06-18 (Streamable HTTP transport) and negotiates down to 2025-03-26 or 2024-11-05 if the client requests an older version. Most current Claude clients should work without manual version pinning.

Yes — any MCP client that supports remote HTTP servers can connect. ChatGPT Desktop (custom connectors), Cursor (MCP settings), and Claude Desktop (config file) all work. The install command varies by client; the URL is always https://stackswap.ai/api/mcp.

The server is part of the closed-source stackswap.ai codebase, but the public docs repo at github.com/StonesofCreation/stackswap-mcp contains the protocol surface, JSON schemas for all eight tools, an example client, and an open issue tracker. If you want a tool added, removed, or behavior changed, file a GitHub issue there or email nick@stackswap.ai. The MCP surface is shaped by what operators actually ask about.

Endpoints

JSON-RPC: POST https://stackswap.ai/api/mcp
Server descriptor: GET https://stackswap.ai/api/mcp
Protocol versions supported: 2025-06-18, 2025-03-26, 2024-11-05
Auth: none
Rate limit: none (reasonable use)
Source: docs + schemas + example client at StonesofCreation/stackswap-mcp (MIT). The MCP server implementation is closed-source.

Issues, requests, or new tool ideas: nick@stackswap.ai.