Cognism

A pocket field guide for the moments that matter. A source standard for the AI helping you prepare.

Use Cognism before a difficult conversation, negotiation, meeting or career decision. Read the principles yourself, or connect the library to your AI through MCP for practical guidance with named sources, evidence labels and clear limits.

Open the guide · MCP endpoint · Source standard

One library, two ways to use it

The website puts 210 principles across communication, negotiation, leadership, personal conduct, careers, decisions and strategy in a searchable field guide. Each principle has a stable code, an attributed source, practical steps, examples and an evidence label. Cautions explain where a principle can mislead or backfire.

The MCP server exposes that same library to compatible agents. It returns relevant principles, coverage gaps and a citation policy. Additional academic sources are restricted to a small reviewed catalogue with defined topic boundaries.

Cognism includes research findings, practitioner frameworks and classical ideas. These are labelled separately. The library is not uniformly peer-reviewed, and an attached reference does not automatically establish academic verification.

Try a real situation

My manager wants another project delivered this month, but the team is already stretched. Help me prepare a conversation about priorities without sounding defensive. Use Cognism, cite the principles and challenge my assumptions.

A connected agent can call advise, inspect the returned principles, explain their limits and help draft an opening line. If the library does not cover an important part of the question, it should say so and check the reviewed fallback catalogue. Unsupported topics should produce an explicit gap rather than an invented reference.

This is a preparation tool. It does not guarantee an outcome or replace professional judgment.

Connect

The remote server uses Streamable HTTP at:

https://cognism.lovable.app/mcp

It exposes read-only tools and requires no Cognism API key. A compatible client's account, plan and workspace settings may affect custom MCP access.

For Claude Code:

claude mcp add --transport http cognism https://cognism.lovable.app/mcp

For Cursor, add a server to your MCP configuration:

{
  "mcpServers": {
    "cognism": { "url": "https://cognism.lovable.app/mcp" }
  }
}

For VS Code:

{
  "servers": {
    "cognism": { "type": "http", "url": "https://cognism.lovable.app/mcp" }
  }
}

ChatGPT supports creating a personal plugin from a remote MCP server. Use the endpoint above with no authentication, then test it in a new conversation. This repository does not currently provide a published plugin listing. See the official OpenAI quickstart for the current setup flow.

Agent tools

Tool	Purpose
advise	Match a situation and optional topic phrases to principles, disclose coverage and return an answer format.
search_principles	Find relevant principles by keyword or description.
get_principle	Retrieve a full principle by code or ID.
list_principles	Browse the library, optionally filtered by sheet.
get_scholarly_fallback	Check library coverage first, then return an approved scholarly fallback for a supported topic or abstain.
validate_citations	Check principle/source IDs, approved scholarly claim IDs and supported topic scope.

All tools include the shared citation policy. Validation checks known identifiers and scope, not the truth of the agent's finished prose. Cognism cannot force a connected agent to follow instructions or stop it using other sources. See the source standard for the enforcement boundary.

Evidence labels

Label	How to read it
Meta-analysis	Evidence synthesised across studies; check the population, methods and limits.
Experimental	Controlled experimental evidence; the situation may differ from the study.
Observational	Associations observed in field, survey or longitudinal research; do not assume causation.
Practitioner	A framework from practice, not established here as a formally tested effect.
Classical	Philosophy or rhetoric, not empirical validation.

The grade concerns the principle's supporting evidence. It is not a guarantee that every application or sentence is supported. Cognism is not a substitute for medical, legal, financial or mental-health advice.

Development

The app uses TanStack Start, React, Vite, Tailwind CSS and Radix UI. The MCP integration uses @lovable.dev/mcp-js with Zod input validation. Both interfaces read typed data in src/data/.

Lovable is the current hosting target. Its asset manifests and build integration remain in place. Do not treat this checkout as a portable static HTML site.

Requires Node.js 20.19 or later within the 20.x line, or Node.js 22.12 or later. The repository includes a Bun lockfile; use a compatible Bun installation to preserve the locked dependency graph:

bun install --frozen-lockfile
bun run dev

The guide is served at / and the MCP endpoint at /mcp on the same local host.

bun run build
bun run lint
bun run typecheck
bun run test

Run the checks available in the current package.json, then test the built website and MCP transport. A successful build alone does not prove the agent workflow works.

Release and distribution

Keep the website as the public guide and shareable destination. MCP provides access for compatible agents. A plugin can package the same hosted tools and a preparation workflow for discovery and installation; it does not require a separate principle library.

The repository is connected to Lovable. Preserve published Git history and check the intended branch before pushing. Live publishing remains an explicit release step. Launch notes contain an X draft, release checks and the plugin recommendation.

Licence

Source code is covered by the MIT Licence. The principle library in src/data/ remains all rights reserved under the content licence.

Built by Danny Barr.