---

title: InterOrdra MCP emoji: 🔍 colorFrom: purple colorTo: blue sdk: docker pinned: false license: mit short_description: Semantic gap detection tool for AI agents

InterOrdra MCP

smithery badge License: MIT

Measure semantic distance between texts. Detect misalignment before it becomes a problem.

InterOrdra is an MCP server that tells you — with a number — how far apart two pieces of text are conceptually. Not just keyword overlap: real semantic distance using embeddings.

Built for AI agents, pipelines, and developers who need to know when a conversation, a retrieval, or a response is failing silently.

---

The problem it solves

Two things can be syntactically connected but semantically worlds apart:

• A user asks about X. Your agent responds about Y. Nobody notices.
• Your RAG pipeline retrieves documents. They don't actually answer the query. The LLM hallucinates to fill the gap.
• A negotiation goes on for hours. The parties are never talking about the same thing.
• A user rephrases the same question 5 times. The system keeps missing the real need.

InterOrdra surfaces these gaps. It gives you a score from 0 (fully aligned) to 1 (completely disconnected), the severity level, and the vocabulary unique to each side.

---

Connect via Smithery (recommended)

No setup required. Works with any MCP-compatible client.

Connect on Smithery

You'll need your own ANTHROPIC_API_KEY. Smithery will prompt you for it on connect.

---

Use cases

Scenario	Tool to call
Check if an LLM answer is relevant to the question	detectar_gap
Validate RAG retrieval — does the doc actually answer the query?	detectar_gap
Two agents in a pipeline producing disconnected outputs	detectar_gap
User keeps rephrasing the same question unsatisfied	reformular_pregunta
Find the real need behind a vague request	reformular_pregunta
Multi-turn conversation drifting and losing coherence	analizar_conversacion
Diagnose why a negotiation or discussion failed	analizar_conversacion
Detect misalignment between two team members' messages	analizar_conversacion

---

When to call each tool

Call detectar_gap when:

• A question and its answer seem off-topic or disconnected
• You need a numeric score for semantic similarity between two texts
• You're building a relevance filter for retrieval-augmented generation
• Two concepts need to be verified as belonging to the same semantic space

Call reformular_pregunta when:

• A question is too vague to answer well
• A user keeps asking the same thing without getting satisfaction
• You need to surface the underlying problem before responding

Call analizar_conversacion when:

• A multi-turn conversation is drifting and losing coherence
• You need to find the exact turn where alignment broke down
• An agent pipeline is producing inconsistent outputs across turns

---

Tools

detectar_gap

Measures semantic distance between two texts using embeddings. Returns a gap score, severity level, and the vocabulary unique to each text.

Input:

{
  "texto_a": "the server is not responding to network requests",
  "texto_b": "I need the team to understand my product vision"
}

Returns:

{
  "gap_score": 0.94,
  "nivel": "alto",
  "mensaje": "Gap semántico significativo. Los textos hablan de mundos distintos.",
  "similaridad_semantica": 0.06,
  "palabras_solo_en_A": ["servidor", "red", "solicitudes"],
  "palabras_solo_en_B": ["visión", "producto", "equipo"],
  "metodo": "embeddings"
}

Gap score:

• 0.0 – 0.3 → Low. Texts share enough meaning.
• 0.3 – 0.6 → Medium. Partial disconnection. Misunderstandings likely.
• 0.6 – 1.0 → High. Texts operate in completely different conceptual worlds.

---

reformular_pregunta

Takes a question and returns three alternative framings that surface the real need behind it. Uses Claude.

Input:

{
  "pregunta": "why doesn't anyone understand me"
}

Returns:

{
  "pregunta_original": "why doesn't anyone understand me",
  "variantes": [
    "What specific communication breakdown is happening in your current context?",
    "What would it look like if someone truly understood you — what would change?",
    "Which part of your message consistently gets lost or misinterpreted?"
  ],
  "instruccion": "Use these variants to explore the gap between what is asked and what is needed."
}

---

analizar_conversacion

Analyzes a sequence of messages to detect accumulating semantic gaps. Finds where a conversation starts drifting apart.

Input:

{
  "mensajes": [
    "We need to improve system performance",
    "I think we should hire more engineers",
    "The budget for Q3 is already allocated",
    "Can we talk about team morale instead?"
  ]
}

Returns:

{
  "gaps_detectados": [
    {"entre_mensajes": "1 y 2", "gap_score": 0.45, "nivel": "medio"},
    {"entre_mensajes": "2 y 3", "gap_score": 0.71, "nivel": "alto"},
    {"entre_mensajes": "3 y 4", "gap_score": 0.83, "nivel": "alto"}
  ],
  "gap_promedio": 0.66,
  "punto_critico": {"entre_mensajes": "3 y 4", "gap_score": 0.83},
  "diagnostico": "Conversación gravemente desacoplada"
}

---

Self-host

Requirements: Python 3.10+ · Your own ANTHROPIC_API_KEY

pip install fastmcp anthropic
python server.py

Claude Desktop — add to claude_desktop_config.json:

{
  "mcpServers": {
    "interordra": {
      "command": "python",
      "args": ["/path/to/server.py"],
      "env": {
        "ANTHROPIC_API_KEY": "your-api-key-here"
      }
    }
  }
}

InterOrdra uses your own Anthropic API key. The author does not pay for your usage.

---

Background

InterOrdra emerged from a pattern: two systems broadcasting on completely different frequencies — technically communicating, actually disconnected.

The name comes from inter (between) + ordra (order/structure) — the space between ordered systems where gaps live.

Full project: github.com/rosibis-piedra/interordra

---

Author

Rosibis Piedra AI Software Engineer · Costa Rica github.com/rosibis-piedra

---

License

MIT