Skip to main content
Glama

ayon-mcp

A lean, standardized MCP server for AYON (Ynput) — drive an AYON production server from any AI agent through one clean interface.

Fourth in a set of single-purpose tracker MCPs, all the same shape so an agent (or a migration) can speak to any of them interchangeably:

shotgrid-mcp · ftrack-mcp · kitsu-mcp · ayon-mcp

What it gives you (20 tools)

  • One generic CRUD family over AYON's entities — find / get / create / update / delete (entity types: folder, task, product, version, representation).

  • Schema / discoverylist_projects, get_project, list_folder_types, list_task_types, list_statuses (with the canonical mapping), list_tags, get_attributes, list_addons, whoami.

  • Typed helpersnew_folder (any folder type), new_task, new_product, new_version, set_status.

  • project_summary — a normalized, cross-tracker snapshot (counts + per-shot tasks with canonical statuses) in the exact same shape the other three MCPs emit, so hub tools (verify / migrate / audit) work on AYON for free. AYON's polymorphic folders are mapped onto sequences/assets/shots by folder_type.

Related MCP server: MCP Server Template

What's standardized here (vs a raw AYON client)

  1. One CRUD family instead of dozens of typed endpoints — find("folder", project, {...}) etc.

  2. A two-level dry_run on every write"plan" (client-side echo, contacts nothing) and "preflight" (resolves references + validates statuses against live data, returns a before→after diff and an ok/would_fail verdict — writes nothing). Optional MCP_PLAN_LOG=/path.jsonl records every plan.

  3. Canonical statuses — AYON's per-project statuses (Not ready / In progress / Pending review / Approved…) are mapped to the shared todo/wip/done/review/approved set, so cross-tracker logic is uniform.

  4. The normalized project_summary contract — identical to shotgrid/ftrack/kitsu-mcp.

Install

pip install ayon-python-api fastmcp

Configure (env only — no secrets in source)

export AYON_SERVER_URL="http://your-ayon:5000"
export AYON_API_KEY="<a service / API key>"      # create one in AYON ▸ user ▸ API keys

Add to Claude Code / any MCP client:

claude mcp add ayon \
  -e AYON_SERVER_URL=$AYON_SERVER_URL \
  -e AYON_API_KEY=$AYON_API_KEY \
  -- python /path/to/ayon-mcp/server.py

Part of a tracker-MCP set — migrate between platforms

Because all five MCPs emit the same project_summary and accept a uniform tool surface, an agent with two loaded can copy a project across trackers (read source → write target) with no bespoke script — the clone-as-hub thesis as five shippable MCPs. AYON's own ayon-ftrack addon does exactly this kind of sync internally; this MCP exposes AYON to the same agent-driven workflow.

AYON specifics handled

  • Polymorphic folder hierarchy — folders are Episode/Sequence/Shot/Asset (any nesting); new_folder takes a folder_type, and project_summary flattens them onto the cross-tracker shape.

  • Product → Version → Representation publish model — first-class in the CRUD family.

  • Anatomy / attributesget_attributes(entity_type) + get_project expose the schema-as-data.

Built on the official ayon-python-api (ayon_api) + fastmcp. MIT. Credits Ynput for AYON (AGPL server, open source).

Docs

📊 COMPARISON.md — side-by-side of the five trackers (ShotGrid · ftrack · Kitsu · AYON · NIM): data model, status vocabularies, and the migration incompatibilities to know about.

🧪 TESTING.md — how these servers are validated (live round-trip tests + two-level dry-run checks).


Built by John Huikku · alienrobot.com

Related MCP Connectors

Related MCP Servers

  • F
    license
    Not graded
    quality
    D
    maintenance
    A production-ready Python scaffold for building Model Context Protocol (MCP) servers using FastMCP. It provides a structured framework for developers and AI agents to rapidly develop, test, and manage custom tools and workflows.
    1
    -
  • A
    license
    Not graded
    quality
    D
    maintenance
    A lean MCP server giving LLM agents full access to the ShotGrid / Autodesk Flow Production Tracking API through 15 curated tools, including generic CRUD, schema discovery, and safe writes with a dry_run flag.
    1
    MIT