MCP server

Deterministic conversation dynamics for one thread: message balance, initiation, question rate, average lengths, and who is waiting on whom. Use before advising on next steps.

Search discovery candidates with filters. Filters: minSignal (0-100), maxDistanceMiles, intentAlignedOnly, sharedInterestsOnly.

What is missing from the acting user's profile and MindSync assessment, plus lowest-confidence traits. Use this before coaching the user to complete their profile - then ask conversationally and submit via update_my_profile / submit_assessment_responses.

Full message thread between the acting user and one counterpart.

Today's curated set of candidate profiles for the acting user, ordered by compatibility signal.

Explain why a target user is (or is not) a strong match: signal percent, supporting reasons, friction points, and conversation starters.

Full profile of the acting user: photos, prompts, interests, verification, and location state.

The acting user's complete psychometric analysis: Big Five with facets, emotional intelligence, relationship profile (attachment, love languages, communication/conflict styles), values, confidence per trait, and archetype summary.

The next MindSync questions for this user (server-targeted batch first, then any unanswered). Returns full text and options so you can ask them naturally in conversation.

Shared profile of another user by id. Visibility follows platform rules exactly as in the app.

How the acting user's traits moved over a window: changed, stable, and confidence shifts.

Psychometric summary of another user. The platform only returns what that user permits; expect NOT_FOUND or FORBIDDEN for non-shared profiles.

Acting user's trait trajectory over time from repeated assessment rounds. Optional traitNames filter, daysBack window (default 30).

Like a candidate on behalf of the acting user. If the target already liked the viewer this creates a mutual match; otherwise it registers interest (pending).

Inbox summary of active conversations: counterpart, unread count, last message, block states.

All matches for the acting user with status and compatibility score.

Seed clearly-labeled SYNTHETIC persona profiles (BrandBrain-generated demo fixtures) directly into Postgres, fully onboarded in one call. Beyond the profiles row this writes, in the same transaction: photo captions, a psychometric_analyses snapshot (MindSync-shaped), a verified user_kyc_state row, and match preferences in lifestyle_preferences — so personas read as complete, verified members. Rows are forced to the internal_test audience (never public), emails use persona-<slug>@seed.psymatch.me, and all derived data is deterministic per slug. Re-seeding is idempotent (existing analyses are kept; KYC is refreshed). Requires PSYMATCH_MCP_DATABASE_URL (or DATABASE_URL) in the MCP server env. Use dryRun first, then seed, then verify with admin_list_users (profileAudience=internal_test). This tool fully replaces scripts/personas/seed-personas-supabase.ts and scripts/personas/enrich-personas-supabase.ts.

Submit answers the user gave you for MindSync questions. Each item needs questionId + answer; choice questions accept an option value/id/text, scales accept a number. Ask naturally, confirm wording with the user for open text, then submit.

Profile-aware opener suggestions for starting a conversation with a match.

Update profile text fields on behalf of the acting user. Only send content the user has seen and approved word-for-word. Omitted fields are left unchanged.

Identify the acting PsiMatch user. Use first to confirm whose data you are operating on.

Mark a single message as read on behalf of the acting user.

Accept or reject a pending match. status must be one of ACCEPTED or REJECTED.

Send a chat message to a matched user on behalf of the acting user. Requires agent-send opt-in; rate limited to 10 sends/minute.

Impersonate a user by profile id: every subsequent tool call runs as that user until admin_clear_actor. Requires an admin session; the API rejects non-admins and other admins.

Stop impersonating; subsequent calls run as the real admin again.

Search users as an admin (email/name substring, cohort filter, moderation flags). Use to find profile ids for simulations.

Admin identity probe: the real authenticated admin, whether the API recognizes admin authority, and the current impersonation target.