PlugWiki

Fivetran MCP, the server from Fivetran's GitHub

Official · by FivetranLocal

Lets your AI assistant check Fivetran connection syncs, failures and destinations, and optionally manage them, with your API key.

This is a server from Fivetran's own GitHub organization that you run on your own computer with your Fivetran API key. It lets Claude, Cursor or Codex answer questions like which connections are failing or when Salesforce last synced. Out of the box it can only read; you choose whether it may also change or delete things. It is self-hosted with no support promise in the README, and the install command must point at Fivetran's repo, because the same name on PyPI is someone else's project.

Good for

  • Data engineers who want to ask which connections are broken or behind
  • Analytics teams checking when a source last synced before trusting a dashboard
  • Platform teams creating or copying many connections from Claude Code or Codex
  • Admins auditing destinations, groups and users across a Fivetran account

Works in

4 supported · 8 should work

Tap an app to see how to set it up and any known problems.

Install Fivetran MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the Fivetran MCP server to Claude Code. Run this command:
    
    claude mcp add fivetran \
      --env FIVETRAN_API_KEY=your-api-key \
      --env FIVETRAN_API_SECRET=your-api-secret \
      --env FIVETRAN_SCOPE=read \
      --env DISALLOWED_ACTIONS=system-keys:read,users:read:get_user_api_key,users:read:list_api_keys,users:write:create_user_api_key,users:write:rotate_user_api_key,users:delete:delete_user_api_keys \
      -- uvx --from git+https://github.com/fivetran/fivetran-mcp fivetran-mcp
    
    Ask me for FIVETRAN_API_KEY, FIVETRAN_API_SECRET and only put the values in that config, nowhere else.
    
    Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires uv (Python 3.10 or later), Fivetran API key and secret.

Read-only mode: Leave FIVETRAN_SCOPE set to read, which is the default. The assistant then cannot create, change or delete anything.

Do not shorten the command to uvx fivetran-mcp: that name on PyPI belongs to an unrelated project, so the command must point at Fivetran's GitHub repo. Raise FIVETRAN_SCOPE only when you want the assistant to make changes; the server asks the AI to confirm writes and deletes but does not enforce it. Check it with claude mcp list.

From the project README. Source.

Tools

  • Looks up connections, their status and sync history

    Reads connections (your data sources and their syncs) from the Fivetran API (12 API calls). Available at the default read scope.

    Try: “Which connections failed their last sync?”

    Network · connections_read

  • Creates, changes, pauses or syncs connections

    Creates, updates or triggers actions on connections (your data sources and their syncs) through the Fivetran API (18 API calls). Only appears with read/write scope or higher.

    Try: “Pause the Salesforce connection”

    Write · connections_write

  • Deletes connections

    Deletes connections (your data sources and their syncs) through the Fivetran API (5 API calls). Only appears with read/write/delete scope.

    Try: “Delete the old test Postgres connection”

    Delete · connections_delete

  • Looks up destinations

    Reads destinations (the warehouses data loads into) from the Fivetran API (6 API calls). Available at the default read scope.

    Try: “What destinations do we have configured?”

    Network · destinations_read

  • Creates or changes destinations

    Creates, updates or triggers actions on destinations (the warehouses data loads into) through the Fivetran API (5 API calls). Only appears with read/write scope or higher.

    Try: “Update the Snowflake destination's settings”

    Write · destinations_write

  • Deletes destinations

    Deletes destinations (the warehouses data loads into) through the Fivetran API (3 API calls). Only appears with read/write/delete scope.

    Try: “Remove the unused BigQuery destination”

    Delete · destinations_delete

  • Looks up groups

    Reads groups from the Fivetran API (6 API calls). Available at the default read scope.

    Try: “Show me all connections in the Production group”

    Network · groups_read

  • Creates or changes groups

    Creates, updates or triggers actions on groups through the Fivetran API (3 API calls). Only appears with read/write scope or higher.

    Try: “Update one of our groups”

    Write · groups_write

  • Deletes groups

    Deletes groups through the Fivetran API (2 API calls). Only appears with read/write/delete scope.

    Try: “Delete the empty Staging group”

    Delete · groups_delete

  • Looks up transformations

    Reads transformations from the Fivetran API (4 API calls). Available at the default read scope.

    Try: “Which transformations ran today?”

    Network · transformations_read

  • Creates or changes transformations

    Creates, updates or triggers actions on transformations through the Fivetran API (5 API calls). Only appears with read/write scope or higher.

    Try: “Update one of our transformations”

    Write · transformations_write

  • Deletes transformations

    Deletes transformations through the Fivetran API (1 API call). Only appears with read/write/delete scope.

    Try: “Delete the transformation we replaced”

    Delete · transformations_delete

Read from the source code

Pros and cons

Good

  • Published in Fivetran's own GitHub organization, under the MIT license.
  • Read-only by default, with clear steps up to write and delete.
  • Covers the whole Fivetran REST API, from connections and destinations to users and webhooks.
  • You can block single actions, such as anything touching API keys.
  • Setup steps for Claude Desktop, Claude Code, Codex and Cursor.
  • Can also run as a shared server over HTTP for a team.

Watch out

  • No hosted version; you run it yourself with uv and Python 3.10 or later.
  • No sign-in with your Fivetran account yet; you paste an API key and secret.
  • Write and delete confirmations are not enforced by the server.
  • The README makes no support commitment.
  • Easy to install the wrong package, because the PyPI name belongs to someone else.
  • The assistant must look up each API endpoint before calling it, which takes extra steps.

Permissions

medium risk · 4 things to watch

Medium risk

If you raise its scope, the assistant can change or delete connections, destinations and users, and the server only asks the AI to check with you rather than enforcing it. Keep the default read-only scope and add the README's denylist for API key endpoints.

What it can do

Watch out for

  • Confirmations are only advice to the AI

    Write and delete tools tell the assistant to check with you, but the server does not enforce it. Keep the read scope unless you need changes. github.com

  • It can reach API key endpoints

    Within the scope you grant, it can read, create or rotate Fivetran API keys. Add the denylist the README recommends. github.com

  • HTTP mode needs host checks

    Without allowed origins or hosts set, HTTP mode runs without DNS-rebinding protection. Most people should use the default local mode. github.com

  • Install it only from Fivetran's repo

    The fivetran-mcp name on PyPI belongs to an unrelated project. Use the install command above, which points at Fivetran's GitHub repo. github.com

Security advisories

No published security advisories as of Nov 1, 2026.

Technical details
Turning on read-onlyRead-only is the default: leave FIVETRAN_SCOPE unset or set it to read. Write and delete tools only appear with read/write or read/write/delete; DISALLOWED_ACTIONS removes specific groups or endpoints.
Talks toapi.fivetran.com
Sign-in and keysFivetran API key and secret in environment variables of your client config (local mode), or a Basic Authorization header per request (HTTP mode).
Scopes
  • read

  • read/write

  • read/write/delete

Alternatives

Use Fivetran's own server to check and manage Fivetran itself. dbt's server is the pick for the models downstream of your syncs.

ServerOfficialRunsSign-inCan change thingsGitHub stars
For Fivetran
Fivetran MCP ServerThis pageOfficialLocalAPI keyYes26
Other tools for the same job
dbt-mcpOfficial–––614

Questions

Is there an official Fivetran MCP server?

Fivetran's GitHub organization publishes this server, but you host it yourself and the README makes no support promise. In November 2025 Fivetran's blog said it did not yet have an officially supported MCP server; this repo appeared in January 2026.

Is the Fivetran MCP server free?

Yes. It is open source under the MIT license. You need a Fivetran account to create the API key and secret it uses.

Why does `uvx fivetran-mcp` install the wrong thing?

The fivetran-mcp and mcp-fivetran names on PyPI belong to unrelated projects. Use the command in the install box, which points uvx at Fivetran's GitHub repo.

Is it read-only?

Yes, by default. You can raise it to read and write, or read, write and delete, and block single actions you never want it to take.

Can it delete my connections?

Only if you give it the delete scope. Even then, the server only asks the AI to confirm with you first, so keep delete off unless you need it.

Can I sign in with my Fivetran account instead of an API key?

Not yet. Sign-in support is started in the code but does not work, and an open request asks for it. Use an API key and secret.

Is there a hosted Fivetran MCP URL?

No. Fivetran does not host it. You can run it yourself as an HTTP server so several people connect to one copy, and each person then sends their own key.

What can I ask it?

Questions like which connections are failing, when the Salesforce connection last synced, or which destinations you have. With write access it can also create, sync and change connections.

Examples

  • Find failing connections

    What connections are failing?

    Source
  • Check when a source last synced

    When did the Salesforce connection last sync?

    Source
  • List the connections in one group

    Show me all connections in the Production group

    Source
  • Review configured destinations

    What destinations do we have configured?

    Source
  • Copy existing connections to a new destination with Fivetran's copy-connections plugin for Claude Code and Codex

    Source

Information

updated Nov 1, 2026

Publisher
Fivetran (official)
Transport
stdio, Streamable HTTP
Auth
API key
License
MIT
Last commit
Oct 21, 2026

GitHubChangelogIssues

Other Fivetran MCP servers

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 6
  1. fivetran/fivetran-mcp README · github.com
  2. Integrate data faster using natural language, Fivetran, and MCP · fivetran.com
  3. Available actions · github.com
  4. fivetran-mcp server.py · github.com
  5. OAuth Support · github.com
  6. Crash on MCP SDK 2.x · github.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 1, 2026.