Connect your agent

Last updated 2026-06-11 — about 2 minutes with Claude, Claude Code, ChatGPT, or any MCP-capable agent

Perspicuity is a persistent, structured home for real decisions. It walks a decision through the PrOACT method — Problem, Objectives, Alternatives, Consequences, Tradeoffs — and stores the result as a living document you can revisit, share, and act on. There is no chatbot inside: your own agent does the facilitation, connected over the open Model Context Protocol (MCP). You bring the agent; Perspicuity brings the method, the structure, and the memory.

The endpoint. One URL, everywhere: https://perspicuity.ai/mcp (Streamable HTTP). Auth is OAuth 2.1 — you sign in through your browser when you connect. There is no API key to copy and paste.

Did it work? This page will tell you.

Checking your connection status…

Pick your platform

claude.ai Claude Code ChatGPT Any MCP client

On claude.ai (and the Claude desktop and mobile apps, on paid plans):

Open Settings → Connectors.
Choose Add custom connector.
Name it Perspicuity and paste the URL:

https://perspicuity.ai/mcp

Claude opens a browser window; sign in to Perspicuity (or create an account) and approve the consent screen. The OAuth handshake is automatic — that's it.

Then just ask Claude to help you with a decision — "help me decide whether to relocate the team — use Perspicuity" — and it will create and build the decision in your workspace.

From a terminal:

claude mcp add --transport http perspicuity https://perspicuity.ai/mcp

Inside Claude Code, run /mcp and pick perspicuity to authenticate — it opens the same browser sign-in. Claude Code can also hold an encryption key in its runtime, which makes it the natural client for sealed decisions (see Auth and your data below).

Open Settings → Connectors (available on Plus, Pro, Business, and Enterprise plans).
Choose Create (add a custom connector) and paste the MCP server URL:

https://perspicuity.ai/mcp

ChatGPT discovers our OAuth metadata, registers itself as a client automatically, and walks you through sign-in and consent in the browser.

Perspicuity implements the search / fetch contract ChatGPT expects, so it works as a standard connector and inside Deep Research out of the box. For the full authoring loop — creating decisions, adding objectives and alternatives, scoring consequences — enable Settings → Connectors → Advanced → Developer mode, OpenAI's gate for write-capable tools. (Via the OpenAI API and Agents SDK the full tool surface is available with no toggle.)

Any MCP client that speaks Streamable HTTP with OAuth 2.1 discovery connects with zero manual configuration. For clients configured by JSON:

{
  "mcpServers": {
    "perspicuity": {
      "type": "http",
      "url": "https://perspicuity.ai/mcp"
    }
  }
}

Discovery lives at the standard well-known URLs (/.well-known/oauth-authorization-server, /.well-known/oauth-protected-resource), with dynamic client registration (RFC 7591) and PKCE S256.

What your agent can do

Connected agents see a catalog of focused tools — each one a typed operation on your decision workspace. The headline tools:

Tool	What it does
`create_decision`	Start a new decision — creates it in your workspace and lays out the PrOACT step sections.
`list_decisions`	List your existing decisions, newest first, to resume where you left off.
`read_decision`	Read a decision's full structured tree before editing.
`decision_context`	PrOACT grounding for one decision — each section's shape, how to fill it, and a live completeness score.
`step_contract`	The current step's contract: its core question, the probes worth exploring, and what "done enough" looks like.
`step__submit`	The step handshake — records what the user contributed and advances the decision to the next step.
`render_document`	The whole decision rendered as a clean Markdown document, ready to show inline in chat.
`search` / `fetch`	Keyword search across your decisions + fetch one as Markdown (the ChatGPT / Deep-Research contract).

Behind those sit typed write tools for each PrOACT object — objective__create, alternative__create, consequence__set, tradeoff_move__record, recommendation__set, and friends — so the agent edits structure, not free text. Every tool declares read-only / destructive annotations, so your client can show you exactly what a call can touch before it runs.

Auth and your data

OAuth 2.1 with PKCE (S256 only) and dynamic client registration (RFC 7591) — any compliant MCP client can connect with zero manual setup.
Your account's data only. The token your agent receives is scoped to your member identity — it can read and write your decisions and nothing else.
Revocable. Access tokens expire after 30 days and refresh tokens rotate on use; you can see and revoke every token from your profile at any time. We store only a one-way hash of each token.
Your agent's model calls stay yours. Your agent runs on your side and calls its own model provider — we never proxy that inference and never see its API key.
Sealed decisions are zero-knowledge: sealed content is encrypted by your agent before it reaches us, and the key never leaves your side — sealed content is never readable server-side, by us or anyone else.

Troubleshooting

401 / "authentication required" — your token expired or was revoked. Re-authenticate: in claude.ai or ChatGPT, open the connector and reconnect; in Claude Code, run /mcp and authenticate again.
Tool catalog shows up empty or stale — remove the connector and add it again; clients cache the tool list and a reconnect refreshes it.
ChatGPT only shows search and fetch — that's expected outside Developer mode. Enable Settings → Connectors → Advanced → Developer mode for the full authoring tools.
Opening https://perspicuity.ai/mcp in a browser returns an error — also expected: the endpoint speaks JSON-RPC over POST (Streamable HTTP), not GET. Point your MCP client at the URL instead of visiting it.
The connection check above never turns green — the check needs you signed in to this site in this browser, and it flips only after your agent makes its first call. Ask your agent to list your decisions; that first call lights it up.
Still stuck? Email privacy@peopleandplanet.consulting.

What's next

Connected? The quickstart takes you from here to a worked, shareable decision in about ten minutes.