Integrations
Validate before your build agent scaffolds a line of code.
Trigvale ships an MCP server (@trigvale/mcp) that exposes the validation endpoint to any MCP-aware build agent. Wire it into Cursor, Claude Code, Codex, or ChatGPT-with-MCP and the agent can call validate_idea on a raw founder idea before generating a project skeleton.
1. Issue an agent token
Agent tokens are available on the Integrate plan. Issue and revoke them in Settings. Token format is tv_…; the plaintext is shown exactly once at issue time. Treat it like a password.
2. Drop the MCP config into your agent
Same JSON shape across every MCP-aware client. The transport is stdio; no daemon, no port. Your agent runs npx -y @trigvale/mcp on first tool call.
{
"mcpServers": {
"trigvale": {
"command": "npx",
"args": ["-y", "@trigvale/mcp"],
"env": {
"TRIGVALE_AGENT_TOKEN": "tv_..."
}
}
}
}Where the config goes, by client:
| Client | Config location |
|---|---|
| Claude Code | Project's `.mcp.json` (or run `claude mcp add`) |
| Cursor | `~/.cursor/mcp.json` |
| Codex | Codex MCP config (same shape, stdio transport) |
| ChatGPT (with MCP) | Settings → Integrations → MCP servers |
3. Tell the agent to use it
The MCP server exposes one tool today: validate_idea(rawIdea, save?). A minimal system-prompt nudge that flips the behaviour on:
Before scaffolding a new project from a founder idea, call trigvale.validate_idea(rawIdea). Show the verdict + the dominant archetype's structural cautions to the user. Only proceed with scaffolding if the user confirms after reading them.
Make your agent refuse to build until it's validated
Drop this rule into your agent's project config — Cursor's .cursorrules or Claude Code's CLAUDE.md. It gates new scaffolding on a Trigvale verdict instead of letting the agent build on the first idea it's handed.
# Validate before you build (Trigvale)
Before scaffolding a NEW project, app, or service from a founder idea,
call trigvale.validate_idea(rawIdea) first and read the verdict.
- kill -> Do NOT write code. Report the named structural failure and the
weakest dimension. Ask the user to pivot or drop the idea.
- pivot -> Do NOT scaffold yet. Show the pivot direction + prerequisites.
Proceed only if the user confirms after reading them.
- test -> Proceed, but first surface the top assumption to validate and the
archetype's cautions. Build the smallest test of that assumption.
- build -> Proceed. Still show the archetype cautions before scaffolding.
Never skip this for a brand-new idea. Trigvale surfaces the assumptions in the
idea; it does not predict success — the final call stays with the user.Requires the MCP server configured above. The agent now refuses to scaffold on a kill verdict and makes you confirm a pivot before it writes a line of code.
What gets returned
- — A 10-dimension Venture Readiness Score (0–100) with per-dimension breakdown.
- — An evidence brief with each item tagged
observed | inferred | missing | ai | user-claim. - — A deterministic kill / pivot / test / build verdict.Build is rare by construction.
- — Founder-fit calibration deltas if the user has a skill graph on file (re-weights founder-fit, reachability, and execution-complexity dimensions deterministically from declared inputs).
- — 1–3 archetype assignments with structural cautions for the dominant archetype.
- — A vault deep link if
savewas true (default).
What we don't claim
Trigvale exposes assumptions; it does not predict outcomes. The rubric score is one term in a noisy equation dominated by execution, marketing, distribution, timing, and luck. See methodology for the full stance.