Developer room · perspective_room.v1

A room your harness can read in one request.

Every Perspective Room serves three surfaces from one source of truth. No SDK required: one GET and your agent has the company, structured.

GET /r/[token]human room, HTML

GET /r/[token]/agentroom.json payload

GET /r/[token]/agent.mdtext-first brief

Quickstartcopy

$ curl -H "X-Agent-Id: my-harness" \
    https://perspectiveapp.io/r/tk_9f2ma…/agent

{ "schema_version": "perspective_room.v1",
  "company":  { "name": "Meridian Robotics" },
  "story":    { "status": "founder_approved" },
  "materials": [ ... ],
  "boundaries": {
    "diligence_performed": false } }

Connect Perspective to your AI harness

Three ways in, lowest friction first. Room-specific URLs and tokens live in the app under Share. These docs cover the shapes.

01 · Paste the URL

Any chat harness

Give your assistant the /agent.md URL. Text-first, no parsing, works everywhere today.

02 · Fetch the JSON

Pipelines & internal tools

GET /agent with an X-Agent-Id header. Stable IDs, typed fields, last_updated for cheap polling.

03 · MCP server

Claude & MCP-native agents

The MCP package is not published yet. Until then, use /agent.md or /agent directly; the manifest will publish the install command when it is available.

Reads are attributed and logged; founders see your X-Agent-Id label in their activity view. Gated materials return metadata with download_url: null. Respect it.

Reference · the contract

Open source docs

room.json — field referencespec

materials.json — IDs, versions, access statesspec

boundaries — what the payload never assertsspec

agent identity & read logginghosted

Contract items marked spec live in the open-source docs. Hosted items are Perspective-specific.

Reading this with an LLM? Feed it everything:/docs/llms.txt