Skip to content

Rapidly MCP

Test the idea before you build it. Rapidly works inside your AI agent: in about five minutes it takes an idea through a Lean Canvas, turns the riskiest assumption into a measurable hypothesis with a researched pass mark, and returns a build prompt for the smallest Pretotyping experiment worth running. The agent builds the test; real customers provide the evidence. Work is saved to your Rapidly team instead of disappearing in a chat thread.

Server type: Remote Streamable HTTP (hosted)
Endpoint: https://www.rapidly.co/mcp
Auth: one-time bearer token (RAPIDLY_MCP_TOKEN) for new teams, OAuth for existing teams
Tools: 21 (idea generation, Lean Canvas design, hypothesis and experiment design, evidence interpretation, portfolio scoring, team records)
Repo: github.com/Exponentially-Platform/Rapidly-MCP (MIT, 1 star)
Registry: co.rapidly/rapidly
Pricing: free trial with five new ideas; team plans by request

Why This Matters for Operators

Asking a general model for a business plan produces something that looks reasonable and proves nothing. Rapidly encodes the method instead: Alberto Savoia's Pretotyping combined with the vendor's practice across 4,000+ client experiments and 200+ workshops. The core run builds a full Lean Canvas, converts the riskiest demand assumption into a market-engagement hypothesis with a researched pass threshold, then designs three to five named experiments and recommends the smallest useful one with a build prompt. The output includes GO/NO-GO rules, cited sources, and a test asset your agent can build immediately.

Tools & Capabilities

Tool What it does
independent_generate_ideas Generates grounded ideas from a company, problem, or brief with cited public sources
independent_design_lean_canvas Builds a complete Lean Canvas, preserving the original customer, problem, and proposition
independent_design_hypothesis Turns the riskiest assumption into a market engagement hypothesis with a researched pass threshold
independent_design_experiments Creates three to five named Pretotyping experiments, recommends the smallest useful one, returns its build prompt
design_hypothesis Designs a hypothesis and researched target from a saved idea or raw idea text
design_experiments Designs three to five Pretotyping options for a saved idea
interpret_results Turns recorded experiment evidence into an executive summary
evaluate_idea Summarises the evidence and recommends what to do next
score_portfolio Orders the team's ideas by stored value and effort, with confidence and experiment counts
get_sprint_pack Combines the ranked portfolio, team knowledge, focus idea, hypothesis, and fresh experiment designs
list_ideas Lists the team's idea portfolio
get_idea Opens one idea and its current method work
list_experiments Lists an idea's experiments, outcomes, and lessons
get_team_knowledge Reads company context and lessons saved by the team
save_idea Updates selected fields on an existing idea
expand_idea Stages proposed Canvas, hypothesis, and experiment work for review
save_experiment Saves a new experiment with the idea it tests
save_team_knowledge Adds company context or a lesson to team knowledge
create_idea_bundle Creates an idea with optional Canvas, hypothesis, and experiment records
generate_ideas Generates grounded, de-duplicated ideas inside an existing team portfolio
validate_idea Produces a quick Canvas, hypotheses, and a suggested first experiment without public-source grounding

Installation

Open rapidly.co/rapidly-mcp, enter a name, email, and team name. Rapidly creates the account and team, then shows the personal token once (no plan or password needed). The page provides copy-ready setup for Codex CLI and Claude Code.

Configuration

  • New teams: keep the bearer token in the RAPIDLY_MCP_TOKEN environment variable and register the endpoint https://www.rapidly.co/mcp with that header.
  • Existing teams: connect with OAuth, choose the team, and approve access.
  • Works with Claude, Codex, ChatGPT, and Hermes; any agent that supports remote MCP can connect.
  • Rapidly checks team and access scoping on every request.

Business Relevance

  • Pre-build validation: a build prompt for the smallest experiment instead of a full product commitment.
  • Evidence-based decisions: GO/NO-GO rules and pass thresholds reviewed by real customers.
  • Portfolio discipline: score_portfolio ranks ideas by value and effort for funding conversations.
  • Institutional memory: experiments, outcomes, and lessons stay in the team workspace.

Integration with CorpusIQ

Rapidly's experiment evidence pairs with CorpusIQ connectors for the measurement side: validate demand with GA4 landing-page traffic, Shopify trial orders, or Stripe revenue during a Fake Door test, and log hypotheses and pass marks in Notion or Airtable so the validation record lives next to the operating data.

Limitations

  • Free trial caps at five new ideas; team plans require contacting the vendor.
  • Generated facts, cited sources, and proposed targets need human review before use.
  • Bearer token is shown once at creation - store it securely.
  • Young repo (created Aug 24, 2026, 1 star); anonymous endpoint probes return 401 (liveness verified).

See Also

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing