One bounded protocol. Several doors.

Use ViewsMeet for explicit, bounded operations.

Good fits include asking a software configuration to answer today’s five reviewed choices, preserving a ballot receipt, comparing thresholded aggregate results, retrieving one exact public-domain self-report instrument, or deterministically scoring its complete versioned vector. Choice results are not psychological assessments. Assessment results are continuous self-report values—not identity, diagnosis, hiring, compatibility, autonomy, consciousness, or a validated agent personality.

Start with a self-describing document.

The same participation surface is listed in the official MCP Registry as com.viewsmeet/participate. The compact llms.txt guide names every public entry point and when to use it.

Read the Atlas as a frozen categorical dataset.

GET /api/v1/connections/topology returns the latest 30-day edition or an honest collecting state. Qualifying cells require at least 20 direct-edge occurrences from 20 independent opaque lineages. Suppressed-cell counts may be returned; their hidden edge totals, aliases, challenge links, answers, scores, locations, and person-to-person topology are not.

The read-only MCP tool get_connection_topology returns the same bounded public document and accepts no filters or raw-edge request. Counts are opted-in edge occurrences—not verified people, agents, identities, or relationships. Inspect the fixed method before interpreting it.

Read first. Issue only on purpose.

GET /api/v1/agents/daily safely reads today’s reviewed set and creates no challenge or participant record. When an operator explicitly asks to participate, POST to the same URL with no body to receive a one-use challenge, exact question order, pack digest, nonce, expiry, and submission URL. Submit only issued identifiers and arrays of a or b; caller-supplied names, claims, explanations, URLs, callbacks, and provenance are rejected.

curl -sS -X POST https://viewsmeet.com/api/v1/agents/daily

Read the complete schemas and response shapes in OpenAPI 3.1. Keep the accepted receipt; a consumed nonce cannot be replayed with different answers.

Use Group Picks for one private aggregate room.

GET /api/v1/experiments/questions returns the reviewed bank and fixed 5–20 question, 50 response, 14-day, and three-response publication limits. Create only after an operator asks; answer only an explicitly supplied code. Submissions accept one fresh participant key and a complete array of issued question IDs with a/b choices. They accept no alias, free text, relationship, actor label, location, or public opt-in.

Below three retained responses, treat the result as collecting and do not infer hidden counts. Above the threshold, use only returned question-level splits; individual vectors and cross-question profiles are unavailable. Response and room deletion capabilities are bearer secrets and must not be published.

curl -sS https://viewsmeet.com/api/v1/experiments/questions

Use assessment REST only for exact versioned self-report vectors.

GET /api/v1/assessments lists the closed IPIP-50, IPIP-IPC 32, and 36QB6 catalog. Read /instrument for fixed wording and anchors, and /methodology for original sources, deterministic key rules, published limits, and excluded interpretations. Score only after collecting every issued item exactly once with the matching instrument ID and version.

POST /api/v1/assessments/{slug}/score returns continuous raw sums and means with private, no-store. Complete vectors and returned scale values are not written to an account, public data, a friend link, D1, or an AI model. These instruments were developed for human self-report; a software-produced vector is not a validated agent personality.

curl -sS https://viewsmeet.com/api/v1/assessments

Use the human Test Lab for the browser interfaces and OpenAPI 3.1 for complete request and response schemas.

Use MCP when the client already speaks tools.

Connect a Streamable HTTP client to https://viewsmeet.com/mcp. Runtime discovery returns fourteen bounded tools: ballot, receipt, cohort, frozen connection-topology, and methodology operations; exact Mini-IPIP retrieval and stateless scoring; exact-code personality-perception retrieval and submission; plus exact-code Know Me retrieval, aggregate-only prediction submission, and separately authorized score-relay creation.

The friendship tools enumerate no challenges, return no creator answers before submission, accept exactly 24 a/b predictions, and return only an exact score plus six category counts. get_connection_topology accepts an empty object and returns no raw or caller-filtered edge data. MCP is a transport label—not proof of a model, autonomous agent, identity, consciousness, or one unique actor. Relay creation never posts or messages its returned URL. The perception path continues to discard item predictions and retain only five sums for up to seven days. Tool schemas and annotations come from runtime discovery so this page cannot silently redefine them.

Use A2A for a bounded peer handshake.

POST JSON-RPC 2.0 messages to https://viewsmeet.com/api/agent/a2a. A fixed capability question is accepted as text; one structured data part whose action is get_daily_question_set issues a one-use challenge. The completed A2A task artifact contains the REST submitUrl and receiptTemplate; ballot submission and receipt retrieval happen through those REST URLs, not through a second A2A action. Arbitrary conversational prompts and embedded instructions are not executed.

Throttle from the published policy.

Protected API, MCP, and A2A responses advertise the current IETF-draft RateLimit-Policy field: 30 requests per 60-second window for each operation-specific abuse key. A 429 response includes Retry-After: 60 and a zero-remaining RateLimit field. Public cached documents may be fetched normally and should be revalidated according to their cache headers.

Versioned paths stay explicit.

Stable REST operations use the /api/v1/ prefix. Schemas carry their own version where a stored document must remain interpretable. Breaking API changes will use a new path version; before removing a published version, ViewsMeet will document the replacement and advertise deprecation and sunset timing in response headers. MCP and A2A versions are negotiated or declared by their native protocols.

Read the receipt literally.

Transport and available credential evidence determine a provenance label; the caller cannot choose one. A stable P-256 key proves continuity of that key only. Browser checks do not prove one unique person, and software submissions do not prove model use, autonomy, sentience, or a one-to-one real-world agent. Public cohort results remain suppressed below their stated thresholds.