GUIDE, read 10 Nov 2026
Eight facts, all readable from a server's repository or docs in about ten minutes, show what an MCP server for data will do before it touches a warehouse
Start with the publisher and check the claim against the repository
The first line to write down is who publishes the server. For a repository, read the organization that owns it. Fivetran, Collibra, dbt Labs and Sifflet each publish from an organization of their own name. For a hosted server, read which vendor's docs describe it and whether the endpoint is theirs.
Look for a support level. Tableau's README carries a badge reading Tableau Supported. Hex's docs describe an official Hex plugin in the Cursor Marketplace. The community mcp-snowflake-server's README says it is a community project and not published by Snowflake. A label such as Labs is a publisher class and not a promise of upkeep, as the guide on what official means records.
Also check for lookalike packages. Fivetran's README says the PyPI names fivetran-mcp and mcp-fivetran belong to unrelated projects, so it tells readers to run its server from the git URL.
Record the version and the last release as two separate facts
A version names what you install. A last-release date says whether anyone is shipping. Record both and the place you read them. For the dbt MCP Server, GitHub shows v2.10.1 as latest with a release timestamp of 3 Nov 2026, and PyPI shows the same date.
Three gaps are worth writing down as gaps. Hosted servers often publish no version: Select Star's page carries none, and the docs page was last updated seven months before the 14 Jun 2026 archive snapshot that was read. Secoda's repository has no releases, so the date to record is the 26 Jan 2026 commit. ThoughtSpot's releases page shows none, and the npm package's 0.5.0 is dated 6 Jun 2025.
A repository's last push date and a package registry's upload date can differ. For the community Snowflake server the PyPI release is 10 Apr 2025 while the GitHub repository was pushed on 7 Oct 2025, so the directory records both and names the source of each.
Read the transport and the auth route before the config file
The transport decides what you run: a local process over stdio, a hosted URL over Streamable HTTP, or both. The guide on remote versus local servers lists the options. Check whether your client can reach a remote server directly or needs a bridge.
Auth decides whose permissions the agent uses. Look for the word OAuth, then for what happens without it. Per Sigma's help page, the API-key route is labeled Beta. Per Monte Carlo's docs, standard API keys do not work and an MCP-specific key is required. Per Alation's README, the remote server does not support dynamic client registration, so a client_id and client_secret must be supplied. Write down which identity the call runs as; the guide on scoping context covers it.
List the write tools by name and find the control
Open the tool list and search it for create, update, delete, run, deploy, publish, merge, set and put. Then find the page that describes the control. Redshift's README names ACCESS_MODE, DataHub's names TOOLS_IS_MUTATION_ENABLED, Fivetran's names FIVETRAN_SCOPE, and Monte Carlo's names an x-mcp-readonly header.
Where the control is missing from the page, write that down too: Secoda's docs list run_sql and describe no read-only mode (none found on 10 Nov 2026), and Sifflet's README describes no per-call prompt for its two incident-status tools (none found on 10 Nov 2026). Check whether the control is enforced by the server or advised to the client; Fivetran's README says write and delete confirmation is advisory only.
Count the tools and find where the list is published
The count tells you roughly how much of the model's context the definitions take, and the Context Weight Index gives the measured size for the servers it covers. Also note where the list is published, because a count is only checkable if the list is. Euno publishes a table of names and descriptions with no parameter schemas. Sigma's page describes capabilities and publishes no tool names. The Databricks Labs README says tools are inferred at startup from a Unity Catalog schema, and Snowflake's managed server defines tools in a specification that you write.
Source files can be the list. Sifflet's tool names were read from server.py at commit 1f390ca, ThoughtSpot's from tool-definitions.ts at commit 820bf16f6aa3, and Tableau publishes its annotations in source. Where a docs host fails to resolve, say so: the Select Star page was read from a web.archive.org snapshot, and the snapshot date belongs next to the facts.
Runtime requirements, licenses and usage limits belong on the same sheet
A server can pass every check above and still need something the engineer has not got. Fivetran's README requires Python 3.10 or later and uv. Tableau's package needs Node 22.7.5 or later for npx. Collibra's server is a Go binary, so there is nothing to install beyond the release file. Microsoft's README says use is subject to an EULA and NPX installs require explicit acceptance, and ThoughtSpot's npm package is under an end user license agreement.
Usage limits are stated in a few places. Per Airbyte's docs, each organization gets 100,000 direct reads and searches per month at no cost, and other MCP tools do not count toward the limit. Per Collibra's README, natural-language discovery tools use Collibra AI and consume Collibra Units. Google's BigQuery guide says the remote server has no quotas of its own beyond the underlying API quotas.
The dbt Labs README adds a warning to read before install: its dbt CLI tools can modify data models and warehouse objects, so proceed only if you trust the client. Collibra's README says its HTTP mode binds to localhost only and that credentials should be kept in environment variables.
The checklist as a table
The table gives the eight checks, where each is usually found, and an example fact from the directory on 10 Nov 2026. The examples are drawn from different servers and are not a comparison of them.
The examples in the right-hand column come from different servers on purpose. Each shows what the check looks like when the answer is found, and several sections above show what it looks like when the page is silent.
| Check | Where to read it | Example fact |
|---|---|---|
| Publisher | Repository organization; docs site; support badge | Tableau Supported badge on the Tableau MCP README |
| Version | Releases page; package registry; docs stamp | Collibra v0.0.67; Power BI authoring server 1.0.1 |
| Last release | Release timestamp; registry upload date | Redshift server 0.1.3 uploaded 9 Nov 2026 |
| Transport | README client config; docs configuration table | Sigma configuration table sets transport to HTTP |
| Auth | Auth page; environment variables | SYNQ OAuth 2 with PKCE; legacy workspace token |
| Write tools | Tool list; read-only flag or header | Redshift ACCESS_MODE; Monte Carlo x-mcp-readonly |
| Tool count | Tool table or README list | Atlan's page states 41 tools across nine categories |
| Where the list is published | Docs page, source file, or the account specification | Sifflet server.py at commit 1f390ca |
Sources
- Fivetran MCP README, read 10 Nov 2026
- Collibra chip GitHub releases, read 10 Nov 2026
- dbt-mcp GitHub releases, read 10 Nov 2026
- Sifflet MCP server source, read 10 Nov 2026
- Tableau MCP README, read 10 Nov 2026
- Hex MCP server docs, read 10 Nov 2026
- Community mcp-snowflake-server README, read 10 Nov 2026
- PyPI mcp-snowflake-server, read 10 Nov 2026
- Select Star MCP docs (archive snapshot), read 10 Nov 2026
- Secoda MCP GitHub releases, read 10 Nov 2026
- Secoda MCP latest commit, read 10 Nov 2026
- Secoda MCP docs, read 10 Nov 2026
- ThoughtSpot MCP GitHub releases, read 10 Nov 2026
- npm @thoughtspot/mcp-server, read 10 Nov 2026
- ThoughtSpot tool definitions, read 10 Nov 2026
- Sigma MCP server help page, read 10 Nov 2026
- Monte Carlo MCP server docs, read 10 Nov 2026
- Monte Carlo MCP reference, read 10 Nov 2026
- Alation agent SDK README, read 10 Nov 2026
- AWS Labs Redshift MCP README, read 10 Nov 2026
- PyPI awslabs.redshift-mcp-server, read 10 Nov 2026
- DataHub MCP README, read 10 Nov 2026
- Sifflet MCP README, read 10 Nov 2026
- Euno MCP setup docs, read 10 Nov 2026
- Databricks Labs MCP README, read 10 Nov 2026
- Snowflake-managed MCP docs, read 10 Nov 2026
- Atlan MCP tools page, read 10 Nov 2026
- SYNQ MCP docs, read 10 Nov 2026
- npm @microsoft/powerbi-modeling-mcp, read 10 Nov 2026
- Airbyte MCP overview, read 10 Nov 2026
- Collibra MCP README, read 10 Nov 2026
- BigQuery remote MCP guide, read 10 Nov 2026
- dbt MCP README, read 10 Nov 2026
- Power BI Modeling MCP README, read 10 Nov 2026
Last reviewed 10 Nov 2026.