Dubstrata

Model Context Protocol (MCP)

The Dubstrata MCP Server acts as a secure agentic CDN and Graph RAG connector. It bridges Large Language Models (LLMs) to the underlying multi-tenant graph database, real-time market streams, and historical telemetry feeds.

Local Agent Integration

The server runs via Stdio transport and is fully compatible with Claude Desktop, Cursor IDE, and custom agentic frameworks.

"mcpServers": {
  "dubstrata": {
    "command": "npx",
    "args": ["dubstrata-mcp"],
    "env": {
      "DUBSTRATA_API_KEY": "your_tenant_key_here",
      "DUBSTRATA_API_URL": "https://api.dubstrata.com"
    }
  }
}

Exposed Ingestion & Retrieval Tools

1. ingest_knowledge

Ingestion

Ingests new documents and extracts semantic triples (Subject-Predicate-Object) to dynamically register them into the knowledge graph.

Arguments

  • url (string, optional): A URL to crawl, clean of boilerplate code, and ingest.
  • text (string, optional): A raw string payload to parse and ingest directly.
  • tenant_id (string, required): Unique tenant namespace (defaults to "default").
  • is_private (boolean, required): If true, limits the access visibility of this knowledge context to the specified tenant_id.
  • file_path / file_mime / file_bytes_b64 (string, optional): Options for file-upload processing.
  • return_dict (boolean, optional): Returns raw dictionary metadata if true, otherwise returns a summary string.

2. query_graph

Retrieval

Performs a hybrid Graph RAG lookup using vector similarity search, combined with a selective 2-Hop Cypher traversal to gather relevant connections and predictions.

Self-Reflective JIT Ingestion: If the local context is sparse or missing, the tool automatically triggers an asynchronous crawler worker to scrape search engines/wikis for the entity context and update the graph in the background.

Arguments

  • query (string, required): The search query phrase or statement.
  • tenant_id (string, required): Tenant identifier (defaults to "default").
  • is_private (boolean, required): Filter private graph records.
  • agent_role (string, optional): Access role authorization modifier.
  • target_date / domain / relation_types (string/array, optional): Graph search constraints.

Advanced Graph & Analysis Tools

3. get_all_facts

Bypasses the vector search stage to pull every active, latest claim and fact mapped directly to an entity in the graph (up to a limit of 50 records).

Arguments

  • entity_name (string, required): Name of the entity to fetch facts for.
  • tenant_id (string, required): Tenant namespace filter.

4. find_conflicts

Contrasts arguments across multiple sources by grouping claims about an entity by their source URLs. Useful for detecting contradictions or divergent sentiment.

Arguments

  • entity_name (string, required): Target entity name.
  • tenant_id (string, required): Tenant namespace filter.

5. get_historical_timeline

Time machine query that extracts expired or superseded (is_latest = false) claims about an entity to build a historical state change timeline (up to 20 steps).

Arguments

  • entity_name (string, required): Target entity name.

6. get_entities_by_category

Queries nodes by their classified categories (e.g. PERSON, ORGANIZATION, GEOPOLITICAL_EVENT).

Arguments

  • category (string, required): The uppercase category label.
  • name_filter (string, optional): Substring filter to prevent massive record lists.

7. get_interaction_hooks

Fetches actionable next-steps (InteractionHook nodes) mapped to a document URL. If the URL hasn't been crawled, it triggers a JIT ingestion loop.

Arguments

  • url (string, required): Target Document URL.

Evaluation, Feedback & Analysis

8. check_source_trust

Aggregates the historic performance of a specific URL domain by scanning all related document claims, measuring consensus alignment versus fallacy rate to grade source trustworthiness.

Arguments

  • domain (string, required): Target website domain (e.g., "nasa.gov").

9. submit_feedback

Submits tenant feedback (scores 1-5) for a transaction log, which directly recalibrates the query weight constraints and similarity index variables inside the database.

Arguments

  • log_id (string, required): Database log transaction ID.
  • score (integer, required): Feedback score between 1 and 5.
  • reason (string, optional): Explanation feedback string.

10. compile_intelligence_report

Compiles a highly detailed, 10-section formal intelligence report (incorporating BLUF, Outlook, Key Findings, and Causal Matrices) by feeding live Graph RAG context into the synthesis engine.

Arguments

  • query (string, required): The prompt target.
  • tenant_id (string, required): Tenant identifier.
  • return_dict (boolean, optional): Returns structured metadata tables along with the report.