MCP servers that answer with a framework, not a guess
32 tools for Claude, Cursor, Claude Code and OpenCode. Each one runs a real analytical or verification procedure and returns structured output. No model call inside the tool, so the same input gives the same answer every time.
npx -y strategy-mcpThat installs the four free tools — swot_analysis, marketing_mix, lean_canvas, decision_matrix — and speaks MCP over stdio. Works in Claude Desktop, Claude Code, Cursor and OpenCode. Package on npm.
The packs
| Pack | Tools | Price |
|---|---|---|
| StrategyBrain Pro Porter's Five Forces, PESTLE, Ansoff, Jobs-to-be-Done, TAM/SAM/SOM, OKRs, roadmap, brainstorm | 8 | $12 |
| FactCheck MCP Claim verification, source fetching, citation checks, anti-hallucination protocol | 6 | $12 |
| LeadGen MCP Find leads, extract from a page, qualify, save, draft outreach | 6 | $12 |
| Monetize MCP Pricing, packaging, offer ladders, maintenance loop | 8 | $0+ |
All four together, 32 tools: The MCP Toolkit at $29, or $79 with a 30-minute setup call. 30-day refund on every pack.
What actually touches the network
Most MCP listings say "fully local" and stop there. Here is the enumerated version:
| Pack | Network |
|---|---|
| StrategyBrain (12) | none |
| Monetize (8) | none |
| FactCheck (6) | 4 tools → Wikipedia, Semantic Scholar, optional Google Fact Check. cross_verify and anti_hallucination_protocol are local. |
| LeadGen (6) | 2 tools → DuckDuckGo, plus the page you point it at. qualify_lead, save_lead, list_leads, draft_outreach are local. |
26 of 32 tools make zero network calls. The six that do hit free, keyless, read-only public endpoints — no account, no billing, nothing written. When a call fails the tool returns UNVERIFIED instead of inventing an answer.
Hosted endpoints
If you can't install anything locally, the same servers run over Streamable HTTP:
POST https://strategybrain-mcp.factcheck-mcp-worker.workers.dev/mcp
x-api-key: <your key>Write to day1ritterbeckaj@gmail.com for a trial key, or buy a pack and you get your own.
Why deterministic
An agent asked for a SWOT will happily write four plausible quadrants. That output is unfalsifiable — you can't tell a real analysis from a convincing one. These tools compute the cross-mapping, rank the options, score the criteria, and hand back the same structure every run, so the reasoning is inspectable and the result is reproducible.
Source
The free four are MIT-licensed: github.com/Ritterbeckaj/strategy-mcp. Read the code before you run it.