Skip to main content

sg-mcp-server

sg-mcp-server lets an AI assistant answer questions about your Autodesk Flow Production Tracking site. Connect it to an MCP (Model Context Protocol) client, ask in plain language, and the assistant looks up your site's live schema and records instead of guessing at entity types, field names, or status values.

The server gives the client three read-only tools over MCP stdio: sg_schema explores entity types and fields, sg_query runs bounded queries, and sg_get opens one record with its web link. Every query is checked against your site's schema first: a misspelled field name returns a list of valid alternatives and is never silently replaced.

This documentation uses SG for Flow Production Tracking, formerly named Shotgun and ShotGrid.

Security model

sg-mcp-server is read-only: it has no tools that create, change, or delete records. Every request runs as one SG script (ApiUser) and sees what that script's permissions allow. The server does not act as the person asking the question, so anyone who can use the MCP client can read anything the script can read. Give the script access only to the data you intend to expose, and keep its key in your client's private configuration.

The server appends every tool call to a local security audit, ~/.sg-mcp-server/audit.jsonl by default. Each entry records the script name, tool, outcome, duration, and returned record IDs, with a hash of the request instead of its contents. If an audit write fails, the server warns on standard error and still answers, so that call is missing from the audit.

An assistant can still misread a result or answer from incomplete data. In the interface study, no tested model answered every task correctly, so check important answers against the records the tools return.

Installation

You need:

  • Python 3.11 or newer and uv.
  • An MCP client that can start a local stdio server.
  • An SG site reachable from this machine, with a script name and API key that can read the data you want to ask about.

From a checkout of this repository, install the command with the official SG client:

uv tool install '.[official]'

The official extra installs shotgun-api3, the Autodesk Python API the server uses to reach your site. The base package does not include an SG client; without one, the server stops at startup with an error that names the official extra.

After updating the checkout, update the installed command:

uv tool install --reinstall '.[official]'

make install and make reinstall run the same commands. Run sg-mcp-server --version to check the installed version. If your MCP client cannot find sg-mcp-server, give it the full path in the directory that uv tool dir --bin prints.

Quick start

Add the server to your MCP client

Add a stdio server to your client's MCP configuration. This example uses the common mcpServers JSON shape; place it where your client expects MCP servers:

{
  "mcpServers": {
    "sg": {
      "command": "sg-mcp-server",
      "env": {
        "SG_MCP_SITE_URL": "https://yoursite.shotgunstudio.com",
        "SG_MCP_SCRIPT_NAME": "your-script-name",
        "SG_MCP_API_KEY": "your-script-key"
      }
    }
  }
}

Replace the three values with your site's address, script name, and API key. SG_MCP_SITE_URL must be the bare HTTPS site address, with no path, query, or fragment. Restart or reconnect the client so that it starts the server; the server runs until the client disconnects.

The server checks its settings when it starts and stops with a one-line error: message if one is missing or invalid; Troubleshooting lists the causes. It does not contact your site until the first tool call, so a wrong key or an unreachable site appears as an error on your first question, not as a startup failure.

Ask your first question

Ask the assistant:

Which projects can this script see?

The assistant typically calls sg_schema to confirm the Project entity, then sg_query to list Projects, and answers with project names from your site. A list of names means the server is working. The exact answer depends on your site's records and the script's permissions.

Next, ask "What fields are available on Shot?" to explore your site's schema before asking about site-specific fields. Using the server has more example questions, direct tool-call arguments, and every setting.

What the tools do

Tool Use it for
sg_schema List entity types, describe one entity's fields, or search field names and labels.
sg_query Find records with validated filters, sorting, and a row limit, or request one native summary.
sg_get Fetch one record by entity type and numeric ID, including its web URL.

A query returns at most 50 records unless it asks for a different limit, and never more than 200; you can change both limits. When more records match, the result says so and suggests narrowing the filters or requesting a summary, so the assistant does not mistake one page for every match.

Documentation

If you want to... Read
Find example questions and tool-call arguments Using the server
Change the time zone, limits, timeout, audit path, or schema cache Settings
Fix a request that fails or a server that will not start Troubleshooting
Find out where a slow request spends its time Latency diagnostics
Understand the three-tool design and model reliability Interface study
See changes by version Changelog

Development

See the development guide for the required tools, CI, and dependency locks. With those tools installed, run:

make setup
make check

The checks never contact a live SG site.

License

sg-mcp-server is available under the MIT License. It is an independent project and is not affiliated with or endorsed by Autodesk. Flow Production Tracking is a product of Autodesk, Inc.

Metadata

Release files for sg-mcp-server 0.1.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for sg-mcp-server 0.1.0
File Size Uploaded
sg_mcp_server-0.1.0.tar.gz 52.0 kB Details

Built distribution (wheel)

Table of built distributions (wheels) for sg-mcp-server 0.1.0
File Interpreter ABI Platform
sg_mcp_server-0.1.0-py3-none-any.whl Python 3 none any Details

Total release size: 112.0 kB

Release files / sg_mcp_server-0.1.0.tar.gz

Download URL sg_mcp_server-0.1.0.tar.gz
Size 52.0 kB
Tags Source
SHA-256 checksum
How to use checksums
f187eafee3d2ccd45e99a45c6bc31003bc3796da1706d6616137d80986980fc1
BLAKE2b-256 checksum
How to use checksums
9d7cee30be31efd2d8991ec0e98accce4d807d4bbe1398ec8f881b0d409b3ecc
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 12, 2026.

Transparency log

Release files / sg_mcp_server-0.1.0-py3-none-any.whl

Download URL sg_mcp_server-0.1.0-py3-none-any.whl
Size 60.0 kB
Tags Python 3
SHA-256 checksum
How to use checksums
4be5ea665f5519ab85766b9aa515e6ab03a2b1b36e012c090b8b9b5f98d9f91d
BLAKE2b-256 checksum
How to use checksums
7b501c93dee12e56de45d6fa67171903cd03207568cdec7eca026d1d976ba34e
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 12, 2026.

Transparency log

Release history Release notifications | RSS feed

This release

0.1.0 This release

2 release files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page