Atlas MCP Server

Model Context Protocol server providing AI access to Apache Atlas data catalog via Apache Knox.

Features

• Basic authentication — Knox proxies ATLAS_USER / ATLAS_PASS credentials through to Atlas.
• Read-mostly — safe exploration of entities, lineage, types, and glossaries; a small set of write tools for tagging and labeling
• Automatic retries — exponential backoff on transient errors

MCP Tools

Admin / Status

• get_atlas_status() — Server health information
• get_atlas_metrics() — Entity and tag counts by type
• get_atlas_version() — Atlas version

Search

• search_entities(query, type_name?, classification?, limit, offset, exclude_deleted) — Basic search; use '*' to browse all
• fulltext_search(query, limit, offset, exclude_deleted) — Full-text search across attribute values and descriptions
• dsl_search(query, limit, offset) — Atlas DSL queries (e.g. hive_table where db.name="default")
• search_by_classification(classification, entity_type?, limit, offset) — Find all entities with a given tag (e.g. PII)

Entity

• get_entity(guid, ignore_relationships?) — Full entity details including classifications and relationships
• get_entity_by_attribute(type_name, attr_name, attr_value) — Lookup by qualifiedName or other unique attribute
• get_entity_classifications(guid) — Tags applied to an entity
• get_entity_labels(guid) — Free-form labels on an entity
• get_entity_audit(guid, count) — Audit history of attribute and classification changes
• get_entities_bulk(guids) — Fetch multiple entities in one call (comma-separated GUIDs)

Lineage

• get_lineage(guid, direction, depth) — Data lineage graph (INPUT, OUTPUT, or BOTH)
• get_lineage_by_attribute(type_name, attr_name, attr_value, direction, depth) — Lineage without a GUID

Types

• list_entity_types() — All registered entity type names
• list_classification_types() — All classification (tag) type names
• get_entity_type_definition(type_name) — Full attribute schema for an entity type
• get_classification_definition(classification_name) — Attribute schema for a classification type

Glossary

• list_glossaries() — All business glossaries
• list_glossary_terms(glossary_guid?, limit, offset) — Terms in a glossary (or all terms)
• get_glossary_term(term_guid) — Term definition and linked entities
• get_entities_for_glossary_term(term_guid, limit, offset) — Data assets linked to a term

Relationship

• get_relationship(guid) — Details of an entity-to-entity relationship

Write operations

• add_classification_to_entity(guid, classification_name, attributes?) — Apply a tag to an entity
• remove_classification_from_entity(guid, classification_name) — Remove a tag from an entity
• add_labels_to_entity(guid, labels) — Add free-form labels (comma-separated)

Setup

Option 1: Claude Desktop (Local)

1. Clone and install:

 git clone 
 cd Atlas-MCP-Server
 python3 -m venv .venv
 source .venv/bin/activate
 pip install -e .

2. Configure Claude Desktop — edit ~/Library/Application Support/Claude/claude_desktop_config.json:

 {
   "mcpServers": {
     "atlas-mcp-server": {
       "command": "/FULL/PATH/TO/Atlas-MCP-Server/.venv/bin/python",
       "args": ["-m", "atlas_mcp_server.server"],
       "env": {
         "ATLAS_GATEWAY_URL": "https:////cdp-proxy-api/atlas/api/atlas/",
         "ATLAS_USER": "",
         "ATLAS_PASS": ""
       }
     }
   }
 }

3. Restart Claude Desktop and start asking questions about your data catalog.

Option 2: uvx (Cloudera Agent Studio)

{
  "mcpServers": {
    "atlas-mcp-server": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/ecole5/atlas-mcp@main",
        "run-server"
      ],
      "env": {
        "ATLAS_GATEWAY_URL": "https:////cdp-proxy-api/atlas/api/atlas/",
        "ATLAS_USER": "",
        "ATLAS_PASS": ""
      }
    }
  }
}

Configuration

All configuration is done via environment variables.

Connection

Authentication

On CDP ATLAS_USER is the workload username and ATLAS_PASS is the workload password.

TLS / HTTP

CDP URL pattern

The ATLAS_GATEWAY_URL is the full Atlas API path through Knox, for example:

https:////cdp-proxy-api/atlas/api/atlas/

A trailing slash is optional.

Example Queries

Once configured, you can ask Claude things like:

• “What entity types are registered in Atlas?”
• “Find all Hive tables in the default database”
• “Show me the lineage for table default.orders@mycluster”
• “Which entities are tagged as PII?”
• “What does the glossary term ‘Customer ID’ mean?”
• “Show me the audit history for entity guid abc-123”
• “Search for all Kafka topics containing ‘events’”
• “Tag entity abc-123 as Confidential”

License

Apache License 2.0