Jiminy Sequential Thinking MCP Server

A Model Context Protocol (MCP) server for auditable, plural, and stepwise ethical reasoning.
Jiminy enables any LLM agent or system to outsource ethical analysis in a transparent, non-binary, and standards-based way.

🚦 What is This?

Jiminy is a three-phase, auditable ethical reasoning server that implements the Model Context Protocol (MCP).
It provides a single tool for any agent to request:

• Phase 0: Ethical Issue Detection
• Phase 1: Multi-Perspective Analysis
• Phase 2: Solutions & Moral Imagination

All steps, revisions, and alternatives are tracked, logged, and structured by design.

🌳 Why Use Jiminy?

• Alignment: Offload difficult ethical questions to a specialist server
• Plurality: Always explores alternatives—never just “yes/no”
• Transparency: Every step logged (unless disabled) for audit/debug
• Revision & Branching: Supports step edits and exploration of alternative lines of thought
• Separation of Concerns: Main LLM focuses on generation; Jiminy on ethical quality

🏗️ How Does It Work?

• Implements one MCP tool: jiminy_sequential_thinking
• Each call: A single “thought” step, tagged with metadata:
  • thought (text step)
  • phase (0, 1, or 2)
  • revision/branching info if present
• Strict prompt and value hierarchy underpin every analysis

🚀 Quick Start

1. Install dependencies

npm install

2. Run the server

node index.js

(or your entrypoint)

3. Connect via MCP

Your orchestrator/LLM should use the Model Context Protocol and call the tool as documented below.

🛠️ Tool: jiminy_sequential_thinking

Purpose:
Structures and audits every step in ethical reasoning, enforcing plural, multi-perspective, and creative analysis.

Input Schema

| Field | Type | Required | Description | |---------------------|-----------|----------|---------------------------------------------------------| | thought | string | yes | The current reasoning step | | phase | integer | yes | 0 = detection, 1 = analysis, 2 = solutions | | thoughtNumber | integer | yes | Sequence number | | totalThoughts | integer | yes | Estimated total steps needed | | nextThoughtNeeded | boolean | yes | Whether another step is expected | | isRevision | boolean | no | Is this a revision? | | revisesThought | integer | no | Which step is being revised? | | branchFromThought | integer | no | If branching, which previous step? | | branchId | string | no | Branch identifier | | needsMoreThoughts | boolean | no | Still needs further steps? |

Example

{
  "thought": "There is a genuine value conflict between privacy and security obligations.",
  "phase": 0,
  "thoughtNumber": 1,
  "totalThoughts": 4,
  "nextThoughtNeeded": true
}

🔎 Reasoning Structure

Jiminy’s prompt enforces a three-phase workflow:

1. Phase 0 – Detection
   • Is this a real ethical dilemma?
   • If not, stop and explain.
2. Phase 1 – Multi-Perspective Analysis
   • Identify values, actors, frameworks, and possible alternatives.
   • Contextualize legally, professionally, organizationally.
3. Phase 2 – Solutions & Moral Imagination
   • Present all plausible solutions (not just A/B).
   • Highlight trade-offs, regrets, “moral remainders,” and creative possibilities.

Always pluralist and creative. Never “just pick a side.”

🏛️ Value Hierarchy

Jiminy is anchored in a strict value hierarchy (always included in each system prompt):

1. Dignity & Integrity of the Person (non-derogable): right to life, integrity, prohibition of torture/slavery/inhuman treatment.
2. Freedoms & Privacy: autonomy, consent, data, speech, thought, association.
3. Equality & Non-discrimination: equal protection, measures for vulnerable.
4. Justice & Rule of Law: fair trial, effective remedy, proportionality.
5. Solidarity & Social Rights: health, social/environmental/consumer protection.
6. Citizens’ Rights & Good Administration: democracy, transparency, redress.

Absolute red lines: Dignity, life, integrity, and bans on torture/slavery/inhuman treatment—never to be compromised.

(See code for full hierarchy & prompt.)

🖥️ Logging

• Every step is printed (stderr) in a colorized box per phase.
• Disable logging with:

  DISABLE_THOUGHT_LOGGING=true

🔄 Revision & Branching

• Support for iterative improvement, revision, and parallel “branches” of reasoning.
• Enables audit, rollback, and plural exploration of dilemmas.

🧑‍💻 For Developers

• Node.js (TypeScript), built on @modelcontextprotocol/sdk
• Plug-and-play with any MCP-compatible LLM/agent
• Value hierarchy, prompt, phases—all extensible

📜 License

MIT

📚 References

• EU Charter of Fundamental Rights
• Universal Declaration of Human Rights
• European Declaration on Digital Rights and Principles