# Catalog, lineage and observability servers describe data, warehouse servers run SQL, and an agent that needs both connects to both

GUIDE, read 10 Nov 2026

**The short answer.** Metadata servers such as Atlan, Collibra, DataHub, Monte Carlo and Sifflet describe assets, owners, lineage and incidents. Warehouse servers such as Redshift and BigQuery run SQL. The overlap is small: Atlan's query_assets returns up to 100 rows, Secoda has run_sql, and DataHub's get_dataset_queries fetches SQL without running it.

## Metadata servers answer what a table is, who owns it and what it feeds

A metadata server fronts a catalog, a lineage graph or an observability product. Its tools return descriptions, owners, tags, glossary terms, lineage and the state of monitors. Per DataHub's README, agents can discover domains, owners, tags and glossary terms. Per Collibra's README, get_business_term_data traces a business term back to its connected physical data assets. Per Select Star's product page, the server returns descriptions, tags and owners.

Observability servers describe pipeline and data health. Monte Carlo's docs say connected agents can investigate alerts, explore assets and lineage, and create and manage monitors. Sifflet's server lists search_incidents, get_incident_scope_by_issue_number and get_downstream_assets_of_asset. Per dbt Labs' README, get_model_health returns run status, test results and upstream source freshness from dbt Platform.

Lineage depth is a parameter on some servers. Alation's lineage tool takes max_depth (default 10) and, per Alation's guide, is a BETA feature that Alation support must enable on the instance. OpenMetadata's get_entity_lineage reaches up to 10 hops, with includeColumnLineage defaulting to false so the default answer is table level. Monte Carlo's agent_observability toolset adds get_agent_metadata, get_agent_conversation and get_agent_trace, which describe AI agents and not tables.

## Query servers run statements against an engine

A query server's main job is to execute SQL under a database identity. The [Amazon Redshift MCP Server](/servers/awslabs-redshift-mcp-server/) lists execute_query. [BigQuery's remote server](/servers/bigquery-remote-mcp/) lists execute_sql_readonly and execute_sql. The Snowflake-managed server lists a SQL execution tool type, and Databricks documents dbsql. In each, the result depends on the privileges of the calling identity or role, as covered in the guide on [scoping context](/guides/).

These servers publish metadata tools too, but their metadata is what the engine itself knows: schemas, tables, columns and comments. BigQuery's get_table_info returns metadata for BigQuery and BigLake objects and lists no glossary or policy tool, per Google's reference. Redshift's README lists no glossary, ownership or classification tool, and retrieve_skill_file returns a Redshift guide covering metadata and SQL.

Databricks says its dbsql, sandbox and web_search MCPs require the Unity Gateway beta while Genie One MCP is generally available, and that calls to the Genie Agent MCP bypass Unity Gateway and rely on the permissions of the Genie Agent and its data. Snowflake's managed server requires SELECT on a semantic view for Cortex Analyst tools, a privilege check at the engine and not in a catalog.

## Some metadata servers also execute or retrieve SQL

The line between the two kinds is not sharp, so the tools are worth reading by name. Atlan's query_assets runs read-only SQL against a connected warehouse and returns up to 100 rows, and per Atlan's security page tools that execute SQL route it through a read-only validator that permits only SELECT-class queries. Secoda's docs describe run_sql as executing SQL against connected warehouses, and they do not describe a read-only mode (none found on 10 Nov 2026).

Alation's guide says its SQL tools need per-data-product data warehouse credentials configured once. DataHub's get_dataset_queries fetches real SQL queries that reference a dataset or column; it returns query history and does not execute a new statement. Select Star's get_asset_queries returns warehouse queries connected to tables or columns. The dbt MCP Server lists execute_sql and text_to_sql alongside its lineage and Semantic Layer tools.

## How Euno's server fits the split

Euno's docs publish a tool table covering metadata, lineage, EQL, audit, documentation, catalog and benchmark tools. Euno's server gives an agent governed context about the data; it does not run the query, which is the job of the warehouse's own server.

The published facts line up with that statement. No listed tool executes SQL against a warehouse, and per Euno's docs query_to_eql converts a filter without executing it. EQL tools search_resources_eql, count_resources_eql, multi_count_resources_eql and pivot_resources_eql query Euno's metadata graph, not warehouse rows. Euno's site says Euno reads metadata only, and its FAQ says Euno does not query customer data.

Per Euno's October 2026 release notes, pivot and multi-count aggregation tools were added through MCP. An engineer wiring an agent would pair a server like this with a query server, and the page for [MCP servers for Euno](/mcp/euno/) holds the record.

## Freshness differs because the two kinds read different things

Query servers read live engine state at call time. The Redshift README describes metadata queried live from the clusters, and the community mcp-snowflake-server reads metadata live from Snowflake on each tool call. Metadata servers read what was collected into the catalog. Per Atlan's security page, metadata is retrieved in real time with no intermediate caching, but the catalog contents depend on Atlan's connectors.

DataHub's README describes no ingestion cadence, so the server reads whatever DataHub has ingested (none found on 10 Nov 2026). The same finding is recorded for the Collibra README, Alation's README and lineage guide, and Select Star's docs page. An engineer checking whether an answer is current should ask when the catalog last collected from the source, because the MCP server does not state it.

## Pairing the two kinds in one agent

An agent often calls a metadata server to find which table holds a metric and who owns it, then calls a query server to read rows. Two servers mean two sets of tool definitions in context, which the [Context Weight Index](/context-weight/) measures, and two identities to check. Databricks documents Genie One as answering natural-language analytics questions grounded in Genie Ontology through one managed MCP, and Cube's runQuery runs Cube SQL against views of its semantic model.

The table lists a sample in alphabetical order by what each server describes and whether it executes SQL.

| Server | Describes data | Runs or retrieves SQL |
| --- | --- | --- |
| Atlan MCP (hosted) | Assets, glossary, lineage, tags | query_assets, read-only SQL, up to 100 rows |
| BigQuery remote MCP server | Dataset and table metadata | execute_sql_readonly, execute_sql |
| Collibra MCP Server | Glossary, lineage, data quality rules | none found on 10 Nov 2026 |
| DataHub MCP Server | Search, lineage, owners, tags | get_dataset_queries retrieves query history |
| Euno MCP | Governed context, lineage, impact | No listed tool executes SQL |
| Monte Carlo MCP Server | Alerts, monitors, asset lineage | none found on 10 Nov 2026 |
| Secoda MCP | Search, lineage, glossary | run_sql |
| Sifflet MCP Server | Assets, incidents, monitors | none found on 10 Nov 2026 |

## Sources

- DataHub MCP README: https://raw.githubusercontent.com/acryldata/mcp-server-datahub/main/README.md (read 10 Nov 2026)
- Collibra MCP README: https://github.com/collibra/chip/blob/main/README.md (read 10 Nov 2026)
- Select Star MCP product page: https://selectstar.com/product/mcp-for-data (read 10 Nov 2026)
- Select Star MCP docs: https://docs.selectstar.com/features/mcp-server (read 10 Nov 2026)
- Monte Carlo MCP server docs: https://docs.getmontecarlo.com/docs/mcp-server (read 10 Nov 2026)
- Sifflet MCP README: https://github.com/siffletdata/sifflet-mcp/blob/main/README.md (read 10 Nov 2026)
- Sifflet MCP server source: https://github.com/siffletdata/sifflet-mcp/blob/1f390ca06d7919214bfe067991539502fdb4b9fc/src/sifflet_mcp/server.py (read 10 Nov 2026)
- dbt MCP README: https://raw.githubusercontent.com/dbt-labs/dbt-mcp/main/README.md (read 10 Nov 2026)
- AWS Labs Redshift MCP README: https://raw.githubusercontent.com/awslabs/mcp/main/src/redshift-mcp-server/README.md (read 10 Nov 2026)
- BigQuery MCP tools reference: https://docs.cloud.google.com/bigquery/docs/reference/mcp (read 10 Nov 2026)
- Snowflake-managed MCP docs: https://docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-agents-mcp (read 10 Nov 2026)
- Databricks managed MCP page: https://docs.databricks.com/aws/en/generative-ai/mcp/managed-mcp (read 10 Nov 2026)
- Atlan MCP tools page: https://docs.atlan.com/product/capabilities/atlan-ai/references/mcp-tools (read 10 Nov 2026)
- Atlan MCP security page: https://docs.atlan.com/product/capabilities/atlan-ai/references/mcp-security (read 10 Nov 2026)
- Secoda MCP docs: https://docs.secoda.co/features/ai-assistant/secoda-mcp-server.md (read 10 Nov 2026)
- Alation authentication guide: https://github.com/Alation/alation-ai-agent-sdk/blob/main/guides/authentication.md (read 10 Nov 2026)
- Alation agent SDK README: https://github.com/Alation/alation-ai-agent-sdk/blob/main/README.md (read 10 Nov 2026)
- Euno MCP setup docs: https://docs.euno.ai/ai/mcp-setup.md (read 10 Nov 2026)
- Euno October 2026 release notes: https://docs.euno.ai/support-resources/release-note/october-2026.md (read 10 Nov 2026)
- Euno website and FAQ: https://euno.ai/ (read 10 Nov 2026)
- Community mcp-snowflake-server README: https://raw.githubusercontent.com/isaacwasserman/mcp-snowflake-server/main/README.md (read 10 Nov 2026)
- Cube MCP server docs: https://docs.cube.dev/docs/integrations/mcp-server.md (read 10 Nov 2026)

Last reviewed 10 Nov 2026. Canonical: https://datamcpservers.com/guides/metadata-servers-versus-query-servers/
