Skip to content
Data MCP Directory

SITUATION, read 10 Nov 2026

Putting agents on the data this year: check transport, auth, write access and tool-definition size, server by server

The directoryAll situations

An agent that reaches the warehouse, the transformation layer, the BI tool and the catalog loads every server's tool definitions into its context window. The Context Weight Index measures that size from each server's published definitions.

Hosted servers with OAuth sign each person in as themselves; local servers usually run with a token on the user's machine. The table shows which is which.

What to check

CheckThe field or fact that answers it
Hosted endpoint or a process on each laptop?Transport field
Does each call run as the person asking?Auth field
Can the agent change anything?Access field and write tool names
How much context do the tool definitions take?Context weight field and the Context Weight Index

The published facts, for the 30 servers in scope

Filed by data tool, read 10 Nov 2026. Scope: warehouses, transformation, ingestion connectors, bi and analytics, catalogs and metadata, data observability.
ServerTransportAuthAccessContext weight
Airbyte MCP (hosted)Remote (hosted)OAuth (Streamable HTTP) as the signed-in Airbyte user, or an Airbyte application for non-browser clients; tools run with that user's Airbyte role.Configurable870 tokens for 50 tool definitions (50 with a description), measured from the docs page read
Alation AI Agent SDK MCP serverRemote and localOAuth client credentials of a service account (role-limited) or user refresh token; remote MCP at <instance>/ai/mcp needs client_id and client_secret (no dynamic client registration).Not stated in the source read69 tokens for 4 tool definitions (3 with a description), measured from the docs page read
Amazon Redshift MCP Server (AWS Labs)Local (stdio)AWS credentials (profile/env) with IAM and Redshift Data API; database user privileges.Configurable238 tokens for 8 tool definitions (8 with a description), measured from the README read
Atlan MCP ServerRemote (hosted)OAuth per user (calls run as the signed-in user) or Atlan API key as Bearer token (shared service identity); local server uses API key + base URL.Configurable1,770 tokens for 42 tool definitions (42 with a description), measured from the docs page read
BigQuery remote MCP serverRemote (hosted)OAuth 2.0 with IAM (Google Cloud identities, OAuth client, or agent identity).Configurable1,487 tokens for 9 tool definitions (9 with a description), measured from the docs page read
MCP Toolbox for Databases (BigQuery tools)Remote and localApplication Default Credentials, client OAuth token forwarding, or service account impersonation.Configurable156 tokens for 9 tool definitions (9 with a description), measured from the docs page read
Collibra MCP Server (chip)Remote and localCollibra username/password: server-wide credentials over stdio (mcp.yaml or env vars), or client-supplied Basic Auth headers per request over HTTP; tools respect the connecting user's DGC scopes.Configurable2,914 tokens for 41 tool definitions (41 with a description), measured from the README read
Cube MCP serverRemote (hosted)OAuth Authorization Code + PKCE (scope mcp-agent-access, public client cube-mcp-client), or a Cube API key exchanged for a token (client_id cube-api-key).Read and write1,093 tokens for 30 tool definitions (30 with a description), measured from the docs page read
Databricks Labs MCP serversRemote and localDatabricks CLI-style credentials; Unity Catalog permissions.Not stated in the source readNot measured: README says the tool list is inferred at startup from the Unity Catalog functions, vector search indexes and Genie spaces the user configures, so no fixed per-tool list or descriptions are published.
Databricks managed MCP serversRemote (hosted)Databricks workspace sign-in; caller's Unity Catalog permissions; Unity Gateway grants and policies.Configurable94 tokens for 4 tool definitions (4 with a description), measured from the docs page read
DataHub MCP ServerRemote and localStdio: DATAHUB_GMS_URL + DATAHUB_GMS_TOKEN (or ~/.datahubenv); HTTP: per-request DataHub personal access token as Authorization Bearer, no shared server token.Configurable567 tokens for 22 tool definitions (22 with a description), measured from the README read
dbt MCP ServerRemote and localdbt service/personal token and environment IDs for platform tools; OAuth or token per docs; local dbt project for CLI tools.Configurable1,061 tokens for 63 tool definitions (63 with a description), measured from the README read
Euno MCPRemote (hosted)OAuth sign-in (default) or API key and headers; active persona scopes data and actions.Configurable760 tokens for 56 tool definitions (42 with a description), measured from the docs page read
Fivetran MCP ServerRemote and localFivetran API key and API secret (FIVETRAN_API_KEY, FIVETRAN_API_SECRET) from the Fivetran dashboard; one shared credential, not per-user OAuth. Streamable HTTP mode forwards headers; OAuth resource-server support is scaffolding only.Configurable299 tokens for 53 tool definitions (2 with a description), measured from the README read
Hex MCP serverRemote (hosted)OAuth authorization to the user's Hex workspace (server URL app.hex.tech/mcp, or the customer's single-tenant, EU or HIPAA domain); requires Explorer role or higher.Read and write252 tokens for 18 tool definitions (18 with a description), measured from the docs page read
Looker tools in MCP Toolbox for DatabasesRemote and localLooker API client_id and client_secret in the Looker source config, or use_client_oauth to pass the caller's OAuth token (default false).Configurable1,475 tokens for 55 tool definitions (55 with a description), measured from the docs index read
Monte Carlo MCP ServerRemote (hosted)OAuth 2.1 with Dynamic Client Registration (user identity, Editor role or above), or MCP-server-type API key id:secret as Bearer token via mcp-remote.ConfigurableNot measured: no published tool-definition text was read on 10 Nov 2026
Omni MCP ServerRemote (hosted)OAuth 2.1 with PKCE (creates an MCP OAuth PAT for the user), or an API key (personal token or Organization API key; Organization keys can act as a user with X-MCP-User-ID).Configurable466 tokens for 27 tool definitions (27 with a description), measured from the docs page read
OpenMetadata MCP ServerRemote (hosted)OAuth 2.0 via existing SSO (per user), personal access token, or Bot JWT as Authorization Bearer.Read and write1,249 tokens for 15 tool definitions (15 with a description), measured from the docs page read
Power BI Modeling MCP ServerRemote and localLocal: Microsoft Entra interactive sign-in (default) or service principal (client secret or certificate). Hosted: Entra ID as the signed-in user.Configurable387 tokens for 21 tool definitions (21 with a description), measured from the README read
Secoda MCPRemote and localSecoda API token (API_TOKEN env var) for the local FastMCP wrapper; docs also describe a Secoda-hosted MCP endpoint authenticated against workspace and AI permissions.Not stated in the source read63 tokens for 6 tool definitions (6 with a description), measured from the source code read
Select Star MCP ServerRemote (hosted)Select Star API token sent as Authorization: Bearer header (mcp-remote proxy used for Claude Desktop).Read only98 tokens for 7 tool definitions (7 with a description), measured from the docs page read
Sifflet MCP ServerLocal (stdio)Sifflet API token (Viewer role for most tools, Editor for monitor generation) and backend URL as environment variables.Read and write662 tokens for 10 tool definitions (10 with a description), measured from the source code read
Sigma MCP serverRemote (hosted)OAuth (HTTP transport) inheriting the user's Sigma account permissions; API-key access token with scope mcp:access (Beta).Read and writeNot measured: The docs page describes features (search, explore, analyze, build workbooks, screenshots, agents) in prose with example prompts but publishes no tool names or per-tool list.
mcp-snowflake-server (community)Local (stdio)Snowflake user credentials (password, private key, external browser) or connections.toml.Configurable103 tokens for 8 tool definitions (8 with a description), measured from the README read
Snowflake Labs MCP ServerRemote and localSnowflake Python connector: password/PAT, key pair, OAuth, SSO, MFA; honors the connection role's RBAC.ConfigurableNot measured: The Snowflake Labs server, marked deprecated in its README, which describes config-driven tool groups (Cortex Search, Analyst, Agent, object management, SQL execution, semantic views) but publishes no per-tool names or descriptions; facts.json lists these six groups.
Snowflake-managed MCP serverRemote (hosted)Snowflake OAuth (default) or External OAuth; PAT supported; RBAC per MCP server and per tool.ConfigurableNot measured: The docs page defines user-configured tool types (CORTEX_AGENT_RUN, CODING_AGENT, CORTEX_SEARCH_SERVICE_QUERY, CORTEX_ANALYST_MESSAGE, SYSTEM_EXECUTE_SQL, GENERIC) used to build tools in a CREATE MCP SERVER specification; it publishes no fixed list of named tools, so facts.json lists tool types not tools.
SYNQ MCP (Coalesce Quality)Remote (hosted)OAuth 2 with PKCE per user with selectable permission tier (read, write, governance, deploy); legacy workspace token (st-...) as Bearer.Configurable4,550 tokens for 134 tool definitions (134 with a description), measured from the docs page read
Tableau MCPRemote and localHosted: OAuth 2.1 with each user's own Tableau Cloud identity. Self-hosted: PAT (default), direct-trust (Connected Apps), OAuth, passthrough, or unified access tokens.ConfigurableNot measured: the source read on 10 Nov 2026 holds the definitions of 2 of the roughly 40 Tableau MCP tools (get-datasource-metadata and query-datasource), and one of those descriptions is imported from a file that was not read, so a measurement would cover a small part of the server.
ThoughtSpot MCP ServerRemote (hosted)OAuth to a ThoughtSpot instance (dynamic client registration, or manually registered client id and secret); bearer-token and /sse paths also exist; the user chooses or is bound to a ThoughtSpot instance URL.Read and writeat least 2,149 tokens for 20 tool definitions (20 with a description), measured from the source code read

What each server documents on clients, auth and scoping

  • Airbyte MCP (hosted): The agent acts on behalf of the Airbyte user who signed in, and every tool runs with that user's Airbyte permissions. Direct reads use the credentials stored in the connector rather than the asking user, so source-side access is not per-user. With the context layer enabled, agents can read live source data, query supported destinations and search synced data by keyword, meaning or both.
  • Alation AI Agent SDK MCP server: The SDK and its MCP server let AI agents use Alation catalog metadata for curation, search and discovery, and data analyst agents. The remote MCP server lists tools dynamically, filtering out tools the current user cannot use. Service-account authentication gives the agent the data and features allowed by the role assigned to the service account. SQL tools need per-data-product data warehouse credentials configured once.
  • Amazon Redshift MCP Server (AWS Labs): The README lists 8 tools for discovery, query and cluster review, scoped to the AWS credentials in use.
  • Atlan MCP Server: The MCP server serves the catalog as a context layer to AI clients (Claude, Cursor, ChatGPT, Gemini, Copilot) and automation platforms. OAuth makes each tool call run as the signed-in user, so tools run with their own permissions and actions are attributed to them. Guests and members are scoped to their persona; MCP narrows but never widens what a user's login permits. The tools page lists 41 tools across nine categories.
  • BigQuery remote MCP server: The server can run queries, get metadata and list resources for AI apps including Claude and ChatGPT. OAuth 2.0 with IAM is used so access follows the authenticated Google Cloud identity.
  • MCP Toolbox for Databases (BigQuery tools): Tools include ask_data_insights (Conversational Analytics API), forecast and analyze_contribution. With useClientOAuth the client's OAuth token is forwarded so calls run as the asking user.
  • Collibra MCP Server (chip): The server is described as giving AI agents access to Collibra Data Governance Center for asset discovery, glossary queries and asset details. Permission scoping follows the connecting Collibra user: tools needing extra DGC scopes are flagged, and a missing scope returns a permission error. Natural-language discovery tools use Collibra AI and consume Collibra Units. With the HTTP transport, MCP clients are advised to supply their own Basic Auth headers per request; with stdio, one server-wide credential is used.
  • Cube MCP server: Every tool runs as the authenticated user, and queries respect Cube permissions including row-level security. With an API key, a token can act as the key's owner, as another user, or be limited to specific tools. Which tools a user sees depends on their role; Viewer role or higher is needed, and data model editing tools show only for roles allowed to edit the model (Admin and Developer by default).
  • Databricks Labs MCP servers: Tools are inferred at startup from the functions and vector search indexes in a given Unity Catalog schema plus listed Genie spaces.
  • Databricks managed MCP servers: System.ai.genie_one_mcp answers natural-language analytics questions grounded in Genie Ontology. The dbsql MCP runs SQL using the caller's Unity Catalog and warehouse permissions, so results are scoped to the asking user.
  • DataHub MCP Server: The server is built to serve DataHub context to AI agents: search, lineage, query history and entity metadata. In the Docker HTTP deployment each MCP client supplies its own DataHub token, so DataHub permissions and audit identity are preserved per user. An optional get_me tool (TOOLS_IS_USER_ENABLED=true) returns the authenticated user's profile and group memberships.
  • dbt MCP Server: The dbt MCP server provides AI agents with context of a project in dbt Core, dbt Fusion and dbt Platform.
  • Euno MCP: Euno MCP uses the active persona; if none can be identified the user is asked to choose. Euno MCP works with Cursor, Claude, ChatGPT in Developer Mode and Claude Code. Euno MCP has a dedicated Gemini Enterprise setup path besides direct, bridge and API-key setups.
  • Fivetran MCP Server: The server uses one Fivetran API key and secret, so agent actions run with that credential's scope instead of the asking user's identity. The README states OAuth resource-server support in streamable-http mode is not yet functional. Two discovery tools plus one execution tool per allowed resource/action pair keep the tool list small and filtered by scope.
  • Hex MCP server: Threads created through the MCP server behave like Threads in the Hex app and select among the data connections the user can access. The MCP server enforces workspace roles and project permissions; editing a cell needs Can Edit or Full Access on that project, even for Admins. Knowledge tools need the Explorer role or above and project editing tools need Editor or above.
  • Looker tools in MCP Toolbox for Databases: The Looker source can pass the caller's OAuth token instead of one service credential via use_client_oauth, which defaults to false. The default source config authenticates with a single client_id and client_secret, so identity is that API user unless OAuth is enabled. The source can hide models, explores and fields from agents with show_hidden_models, show_hidden_explores and show_hidden_fields (all default true).
  • Monte Carlo MCP Server: Monte Carlo says connected agents can investigate alerts, explore assets and lineage, and create and manage monitors. OAuth lets any HTTP-transport MCP client connect with no API keys, and the connector is listed in the Claude Connectors Directory. An Editor role or above is required to use the server. An agent_observability toolset adds get_agent_metadata, get_agent_conversation and get_agent_trace for inspecting AI agents.
  • Omni MCP Server: MCP tools run with the permissions of a specific Omni user, so results include only what that user can access. With OAuth, all MCP queries use the user's Omni permissions and access controls. An Organization API key can run tools as another user via the X-MCP-User-ID header, and X-MCP-Model-ID can pin a model.
  • OpenMetadata MCP Server: Get_user_context returns the authenticated caller's identity, roles, teams, domains and active persona, with no parameter to look up another user. A personal access token gives the agent the same role and access policy as the user. OAuth 2.0 runs tool calls with the user's existing SSO login (Google, Azure, Okta, Auth0, LDAP, SAML).
  • Power BI Modeling MCP Server: The hosted server prompts for Microsoft Entra sign-in on first tool call and then acts with the user's permissions. Fabric IQ, Microsoft's consumption server, answers questions over semantic models and respects permissions, row-level security and object-level security. Microsoft states that clients operate under the user's Fabric RBAC permissions.
  • Secoda MCP: The Secoda-hosted MCP server lets AI tools search, query SQL, read lineage and glossary terms. Authentication verifies the user's workspace and AI permissions, and operations respect governance and access controls. The open-source wrapper authenticates with a single API_TOKEN, i.e. one shared token per server process.
  • Select Star MCP Server: The MCP server is meant to bring catalog metadata and lineage to Cursor, Claude Desktop and VS Code GitHub Copilot. Authentication is a Select Star API token sent as a Bearer header; no per-user OAuth is described, and the docs do not say results are scoped to the asking user. None found: A statement that Select Star MCP results are limited to the asking user's own permissions.
  • Sifflet MCP Server: The README documents setup for Cursor and Claude Desktop through uvx sifflet-mcp@latest.
  • Sigma MCP server: With OAuth the server inherits the user's Sigma account permissions. Access needs the 'Use Sigma MCP with OAuth' account-type permission plus at least Can view on documents or Can use on connections. Agents can call Sigma agents defined in workbooks if the user has Can view on the workbook.
  • mcp-snowflake-server (community): The server exposes a memo://insights resource and an append_insight tool that stores agent-discovered insights. When prefetch is enabled it exposes per-table schema summaries as resources.
  • Snowflake Labs MCP Server: The server served Cortex Search, Cortex Analyst and Cortex Agent services to MCP clients as tools, configured in a YAML file. The server honors the RBAC grants of the role used for the connection, so access is scoped to that Snowflake role.
  • Snowflake-managed MCP server: MCP clients discover and invoke Cortex Analyst, Cortex Search and Cortex Agents as tools on the managed server. Access to the MCP server does not give access to tools; privileges are granted per tool, so results depend on the calling role.
  • SYNQ MCP (Coalesce Quality): OAuth 2 with PKCE means actions are recorded as the real user with a via MCP badge. The set of tools an assistant sees depends on the permissions the user grants at consent, and the consent screen offers only what the user's role allows. The iam tool reports the caller's workspace, granted scopes and user.
  • Tableau MCP: The hosted Tableau MCP at mcp.tableau.com uses OAuth 2.1 so each user signs in as themselves and per-user Tableau permissions are enforced. Authentication to the Tableau REST APIs is selectable between pat (default), direct-trust, oauth and passthrough. The Tableau docs state the authenticated user must have access to the published data sources the agent will use.
  • ThoughtSpot MCP Server: Spotter 'handles authentication securely, and returns live answers from your own data' per the README. The model tools are offered as a group and gated on the user's data-modeling privileges, and get_data is gated on the data-download privilege at runtime. OAuth client registration can be tied to a ThoughtSpot instance so users do not enter the instance URL.

What Euno's server does not do

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.

Other situations

Sources

  1. docs.airbyte.com/platform/airbyte-mcp/tools.md, read 10 Nov 2026
  2. docs.airbyte.com/platform/airbyte-mcp.md, read 10 Nov 2026
  3. github.com/Alation/alation-ai-agent-sdk/blob/main/README.md, read 10 Nov 2026
  4. github.com/Alation/alation-ai-agent-sdk/blob/main/guides/authentication.md, read 10 Nov 2026
  5. raw.githubusercontent.com/awslabs/mcp/main/src/redshift-mcp-server/README.md, read 10 Nov 2026
  6. docs.atlan.com/product/capabilities/atlan-ai/how-tos/remote-mcp-overview, read 10 Nov 2026
  7. docs.atlan.com/product/capabilities/atlan-ai/references/mcp-security, read 10 Nov 2026
  8. docs.atlan.com/product/capabilities/atlan-ai/references/mcp-tools, read 10 Nov 2026
  9. docs.cloud.google.com/bigquery/docs/use-bigquery-mcp, read 10 Nov 2026
  10. raw.githubusercontent.com/googleapis/mcp-toolbox/main/docs/en/integrations/bigquery/prebuilt-configs/bigquery.md, read 10 Nov 2026
  11. raw.githubusercontent.com/googleapis/mcp-toolbox/main/docs/en/integrations/bigquery/source.md, read 10 Nov 2026
  12. github.com/collibra/chip/blob/main/README.md, read 10 Nov 2026
  13. docs.cube.dev/docs/integrations/mcp-server.md, read 10 Nov 2026
  14. raw.githubusercontent.com/databrickslabs/mcp/master/README.md, read 10 Nov 2026
  15. docs.databricks.com/aws/en/generative-ai/mcp/managed-mcp, read 10 Nov 2026
  16. raw.githubusercontent.com/acryldata/mcp-server-datahub/main/README.md, read 10 Nov 2026
  17. raw.githubusercontent.com/dbt-labs/dbt-mcp/main/README.md, read 10 Nov 2026
  18. docs.euno.ai/ai/mcp-setup.md, read 10 Nov 2026
  19. raw.githubusercontent.com/fivetran/fivetran-mcp/main/README.md, read 10 Nov 2026
  20. learn.hex.tech/docs/api-integrations/mcp-server, read 10 Nov 2026
  21. mcp-toolbox.dev/integrations/looker/source, read 10 Nov 2026
  22. docs.getmontecarlo.com/docs/mcp-server, read 10 Nov 2026
  23. docs.getmontecarlo.com/docs/mcp-server-reference-and-operations, read 10 Nov 2026
  24. docs.omni.co/ai/mcp/tools.md, read 10 Nov 2026
  25. docs.omni.co/ai/mcp/authentication.md, read 10 Nov 2026
  26. docs.open-metadata.org/v2.0.x/how-to-guides/mcp/reference.md, read 10 Nov 2026
  27. docs.open-metadata.org/v2.0.x/how-to-guides/mcp.md, read 10 Nov 2026
  28. learn.microsoft.com/en-us/power-bi/developer/mcp/power-bi-authoring-mcp, read 10 Nov 2026
  29. learn.microsoft.com/en-us/power-bi/developer/mcp/mcp-servers-overview, read 10 Nov 2026
  30. raw.githubusercontent.com/microsoft/powerbi-modeling-mcp/main/README.md, read 10 Nov 2026
  31. docs.secoda.co/features/ai-assistant/secoda-mcp-server.md, read 10 Nov 2026
  32. selectstar.com/product/mcp-for-data, read 10 Nov 2026
  33. docs.selectstar.com/features/mcp-server, read 10 Nov 2026
  34. github.com/siffletdata/sifflet-mcp/blob/main/README.md, read 10 Nov 2026
  35. help.sigmacomputing.com/docs/use-sigma-mcp-server, read 10 Nov 2026
  36. raw.githubusercontent.com/isaacwasserman/mcp-snowflake-server/main/README.md, read 10 Nov 2026
  37. raw.githubusercontent.com/Snowflake-Labs/mcp/main/README.md, read 10 Nov 2026
  38. docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-agents-mcp, read 10 Nov 2026
  39. docs.synq.io/scout/mcp.md, read 10 Nov 2026
  40. github.com/tableau/tableau-mcp, read 10 Nov 2026
  41. tableau.github.io/tableau-mcp/docs/configuration/mcp-config/env-vars, read 10 Nov 2026
  42. tableau.github.io/tableau-mcp/docs/configuration/mcp-config/authentication, read 10 Nov 2026
  43. raw.githubusercontent.com/thoughtspot/mcp-server/main/README.md, read 10 Nov 2026
  44. raw.githubusercontent.com/thoughtspot/mcp-server/main/src/servers/tool-definitions.ts, read 10 Nov 2026

Last reviewed 10 Nov 2026.