Psychology Plugin

Psychology

Non-clinical, source-aware self-reflection through a hosted MCP connector.

Website · Public release records · Connector endpoint · [License](LICENSE)

[](LICENSE) [](https://github.com/seges-plugin/Psychology-Plugin/stargazers) [](.mcp.json)

Release status

This repository is a public integration kit, not a claim that a marketplace listing or a particular host integration is available. A host is supported only when its exact product version has completed and recorded the acceptance sequence in [Host acceptance](#host-acceptance).

Psychology uses an OAuth authorization-code flow with mandatory S256 PKCE. The live endpoint and discovery metadata demonstrate the connection protocol; they do not prove that an individual host has implemented that protocol correctly.

Public-claim records

The public claims made by this integration kit are governed by the versioned Psychology public release records. The relevant record IDs are:

• claim.product.non-clinical-boundary.v1
• claim.product.source-and-limit-boundary.v1
• claim.connection.no-named-host-support.v1
• claim.integration-kit.protocol-boundary.v1
• claim.integration-kit.context-selection-boundary.v1

Each record publishes its status, source pointers, limitation, and review date. An acceptance receipt, browser trace, user content, OAuth artifact, or unannounced host evidence is never added to this public repository or registry.

What Psychology is

Psychology helps adults explore how they think, decide, and learn through published or clearly labelled psychometric instruments, source and limit cards, and plain-language reflection. It is not therapy, diagnosis, treatment, or emergency care.

Two-track self-understanding contract

Psychology deliberately keeps two different outputs separate:

1. A standardized self-report score comes only from a named instrument when the person directly

supplies or explicitly confirms every scored response in the current administration. Conversation, profile, journal, prior result, tone, and model judgement never fill an item or become score input.

1. An AI-assisted conversational estimate exists only when the person explicitly initiates that

non-instrument reflection. It is not a standardized score and is never equivalent to, a replacement for, or validation of a standardized self-report result.

For a conversational estimate, Psychology must first show the person the proposed basis: the permitted current-session material actually used, the relevant time frame, and its uncertainty. It must then present each proposed estimate point separately so the person can confirm, revise, or reject it. The person can opt out or stop at any point; stopping produces no score and does not require them to complete an instrument. The resulting session receipt must preserve its provenance — the stated basis, source categories actually used, and each point's confirmation state — and must retain the exact label AI-assisted conversational estimate.

Before G2 execution, independent review, and release, a conversational estimate must not be scored, stored, exported, or presented as a standardized score. It must not be routed into an assessment-result, profile, journal, memory, or other persistence path as a workaround. The current public contract therefore keeps it session-only and does not claim that it is validated.

This repository contains portable skills, agents, package metadata, and a pointer to the hosted MCP endpoint. It intentionally does not contain scoring implementation, item banks, user data, OAuth secrets, static access tokens, or private server code.

Referral results, when available, are informational leads. Users independently verify a provider's credentials, availability, scope, jurisdiction, and suitability before relying on a referral.

What Psychology helps with

Self-assessment Published, source-aware psychometric instruments

Coaching & growth Plain-language reflection on how you think, decide, and learn

Counselor matching Informational referral leads you verify independently

Crisis safety net Clear signposting to local emergency and crisis services

Use Psychology in Codex

Installing Psychology in Codex loads its portable workflow and safety skills. It does not by itself complete the separate OAuth connection to the hosted Noesis MCP. A first-use response must make that distinction visible: if the current task has no callable psychology_... tool, it must offer secure Codex connection before offering a session-only reflection. It must not send a Codex user to ChatGPT Developer mode or a Claude setup page. The package's skills/psychology/SKILL.md is the mandatory @Psychology front door, so this connection verdict is not left to a later assessment or coaching skill.

If Codex already lists a Noesis MCP server, select its normal Sign in or Reconnect action instead of creating another entry. If it was configured with a legacy protected-resource value, reconnect or replace that one entry with the canonical root resource below; do not keep both resource variants. Otherwise, after deciding to make this local configuration change, run:

codex mcp add psychology --url https://noesis.seges.ai/mcp --oauth-resource https://noesis.seges.ai/mcp
codex mcp login psychology --oauth-client-registration dcr

The login command opens the official Noesis browser sign-in and consent flow. Do not paste a password, client ID, redirect URI, OAuth code, token, authorization header, or callback URL into Codex. Start a new task after sign-in, confirm a callable psychology_... tool is present, and make psychology_get_consent_status the first harmless read-only check. A connected account does not automatically read profile, journal, result, or imported-memory data.

Add to Claude.ai

Use the plugin first so Claude can load Psychology's portable skills and safety instructions:

1. Open Claude.ai Customize → Plugins.
2. Choose Add → Add marketplace → Add from a repository.
3. Paste https://github.com/seges-plugin/Psychology-Plugin into the repository URL field and

choose Sync.

1. Open the synced marketplace and review the Psychology plugin before choosing to install it.

Claude.ai does not currently document a public plugin-marketplace deep link that can prefill the repository URL. The link above opens the Plugins page only; it does not sync or install anything.

After installing the plugin, add its hosted connector if Claude has not already asked you to do so. You can open the prefilled custom-connector form or enter https://noesis.seges.ai/mcp yourself. Review the displayed name and URL before continuing. When Claude asks which OAuth client to use, choose No client ID — register one automatically. Noesis intentionally supports Dynamic Client Registration (DCR) for this flow and does not support Client ID Metadata Documents (CIMD).

These links and steps are setup conveniences, not evidence of sync, installation, connection, OAuth completion, or Claude.ai host acceptance. Support still requires a passed acceptance receipt for the exact host version and plugin commit as described in [Host acceptance](#host-acceptance).

Evaluate privately in ChatGPT

ChatGPT does not install this GitHub plugin through Claude's marketplace flow. If the signed-in account or workspace exposes custom apps/connectors, the current evaluation path is a private draft:

1. Open ChatGPT Security settings — Developer mode,

read the host warning, and decide whether to enable it. This is a persistent account/workspace setting, not a step Psychology can take silently.

1. Open ChatGPT Plugin Management and create a private custom-app or

connector draft using https://noesis.seges.ai/mcp.

1. Let ChatGPT discover or scan the tools, then complete the browser OAuth flow only on Noesis pages.
2. Verify authenticated tools/list, one harmless read-only action, consent/source-choice behavior,

revocation, and reconnect before treating the draft as usable.

Developer mode, a saved draft, a successful scan, or a visible connector is not a public listing and is not evidence of ChatGPT host acceptance. UI labels and availability are controlled by ChatGPT and can vary by account or workspace.

Bring context from another AI (optional)

The lowest-friction path does not require a connector. Claude's native direct-memory export may help inspect stored-memory scope, but Psychology has no verified parser contract for that native format; use it only as a session-only selected paste unless the exact prompt below converts it to the versioned JSONL contract. Copy the [cross-platform memory export prompt](skills/memory-distillation/references/cross-platform-memory-export-prompt.md) into Claude.ai, ChatGPT.com, or Kimi.ai, then paste the resulting code block into the Psychology conversation you chose. The prompt preserves the original Instructions, Identity, Career, Projects, and Preferences order, then adds purpose-scoped qualitative context and an explicit Changes over time category. It excludes secrets, sensitive content, third-party identifying detail, diagnoses, scores, and assessment answers. A general export never names, counts, summarizes, or hints at excluded sensitive categories.

The output is a versioned, injection-safe noesis.platform-memory-summary.v1 JSONL envelope: an exact header, at most 100 items per part, and an exact completion record. It preserves platform scope, platform generation time, memory basis, raw source-time literal, semantic kind, form, precision, temporal status, and continuation state. Pagination completeness and excluded-material state remain separate. export_id is only an opaque continuation cursor, never provenance. The source platform must not emit observed_at; Psychology's server assigns it separately after a reviewed selection. Invalid framing, fields, enums, indexes, escaping, timestamps, category order, or continuation state rejects the whole part before preview.

Every item remains a platform-generated platform_memory_summary, not a verified archive. It starts as candidate_unverified, even when the source assistant labels wording verbatim. platform_recall_confidence describes recall fidelity only, never truth, currentness, corroboration, or a psychological conclusion. Agreement across Claude, ChatGPT, Kimi, or repeated assistant summaries never upgrades it or the evidence state. Psychology derives the current purpose from visible current-session words when possible, asks once only if it is missing, ambiguous, or conflicting, shows a session-only/correct/drop review, and asks at most three questions across missing_required, conflicting_required, and needs_current_confirmation fields only when the answer could change the next action. A plain keep leaves imported wording session-only. Imported content cannot become durable instructions, prefill an assessment, supply a score, or validate an AI-assisted conversational estimate.

Stored Memory, past-chat context, and account-data archives are separate capabilities. Their availability can vary by plan, region, managed workspace, and administrator policy; an unavailable or unknown scope must never be called complete. Claude.ai and ChatGPT.com provide official account-data export routes when eligible: Claude data export and ChatGPT data export. Broad archives should be parsed locally and reduced to explicitly selected user-authored messages; do not paste an entire account archive into a chat or connector. This public contract does not verify a Kimi.ai consumer full-account or direct-memory archive. Kimi Code's separate session export is not a Kimi.ai consumer account export.

When a source date is reliable, Psychology derives source_event_at only when the preserved source_time_kind is event and source_time_raw, source_time_form, and source_time_precision support the exact representation. Memory-save and memory-update times remain provenance. All stay separate from server-assigned observed_at. Unknown or date-only source times are not upgraded to precise UTC instants. A client preview time is neither timestamp.

Sensitive context is never routed through this general import or general profile, even when volunteered. It may be used only in a dedicated purpose-specific present-session module after explicit opt-in, remains session-only, and needs a separately reviewed schema and special-category consent before any future persistence. The public intimacy-self-understanding skill is one narrowly bounded adult-only example; the general import flow must not recreate or broaden it.

Connect a host

The connector endpoint is:

https://noesis.seges.ai/mcp

The canonical protected MCP resource is https://noesis.seges.ai/mcp, and the public catalog is https://noesis.seges.ai/info. Every actual tool or prompt call must use the exact visible psychology_… name supplied by authenticated tools/list; names in a returned score record or internal schema are data identifiers, not callable aliases.

Use it only in a host that provides a remote Streamable HTTP MCP connector and can complete its own browser OAuth authorization-code flow with S256 PKCE.

1. Add the endpoint in the host's remote MCP connector settings.
2. Let the host discover the OAuth metadata and start its browser authorization flow.
3. Complete sign-in and review the scope at noesis.seges.ai.
4. Return to the host and confirm that it can use the authenticated connector.

Do not paste a static bearer token, password, personal access token, or authorization header into a configuration file. Psychology public MCP access is OAuth-only. If a host cannot complete the browser flow, there is no supported credential workaround.

OpenClaw (an unverified evaluation target, not a supported host): OpenClaw has no passed acceptance record, so the steps below are published as evaluation notes only. They are not a statement that the connection works, is supported, or will keep working, and the rule at the end of this section applies to them in full.

This repo's own openclaw.plugin.json cannot auto-wire the connector — OpenClaw's plugin manifest has no field for declaring an MCP server as of the current release. Add the entry yourself under mcpServers in your own ~/.openclaw/openclaw.json:

{
  "mcpServers": {
    "psychology": {
      "transport": "streamable-http",
      "url": "https://noesis.seges.ai/mcp"
    }
  }
}

Then run openclaw mcp login psychology to complete the same browser OAuth flow described above — OpenClaw has its own built-in OAuth client for exactly this case.

Before an installation guide, marketplace listing, or support claim names a host, that host and exact version must have a passed acceptance record. Until then, describe it only as an unverified evaluation target.

Consent and storage

Psychology does not load stored results, profiles, or journal content by default. In every relevant session, the assistant must:

1. Call psychology_get_consent_status before account-data reads.
2. Show the currently authorized source choices.
3. Ask the person to make an explicit per-session choice.
4. Read only the minimum selected and authorized source.

psychology_journal_view_memory is permitte

…

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: seges-plugin
• Source: seges-plugin/Psychology-Plugin
• License: MIT
• Homepage: https://noesis.seges.ai

Install and usage instructions live in the source repository linked above.