Creative MCP
Protocol surface · tool schemas

MCP API reference

The Adobe servers expose tools through MCP over stdio. Use the server’s advertised schema as the executable contract; this page explains discovery, calls, instructions, results and where to find exact per-tool parameters.

Transport and lifecycle

The MCP client launches the configured command as a local process and communicates over standard input/output. The process stdout is protocol traffic; startup diagnostics belong on stderr. There is no public HTTP API for the desktop servers.

  1. The client starts the server and negotiates MCP initialization.
  2. The server returns its identity, capabilities and app-specific usage instructions.
  3. The client discovers tool names and JSON input schemas with tools/list.
  4. The client invokes a tool with tools/call and arguments that match the advertised schema.
  5. The server validates inputs, performs the app operation, then returns structured text/content or an error result.

Discover the live API

Do not treat a marketing summary, Atlas row or old copy of a README as the current schema. The connected server’s tools/list is authoritative for the tool names and inputs exposed by that running package.

Discovery

tools/list

Returns advertised tools and their JSON input schemas. Reconnect after a package upgrade so the client refreshes its list.

Detailed guidance

help

App servers with the help utility can describe a tool, list categories, and provide detailed usage/examples. Consult the app-specific guide for its exact formats.

Execution

tools/call

Calls one tool using the exact tool name and schema-defined argument object. Respect enum, unit, target and path constraints.

Client action (conceptual):
  discover tools → choose an advertised name
  call tool with arguments matching that tool's input schema
  inspect isError and returned content before continuing

Use the package tool references for full parameter tables: InDesign generated tool reference, and the Illustrator, Photoshop, After Effects, Premiere and Suite READMEs.

Tool metadata and results

  • Input schema: use the exact property names, required fields, types, ranges and enum values returned by the server.
  • Instructions: server instructions explain app conventions such as units, indices, colors, path rules, errors and safe targeting. MCP clients may pass them to the model as connection context.
  • Annotations: supported tools may identify read-only, destructive, idempotent and open-world behavior. Treat annotations as useful metadata, not a replacement for checking the target.
  • Errors: a failed tool may return isError: true with a message. Stop and interpret that result; do not assume the operation partially or fully succeeded.
  • Stateful application: active document, selection, current sequence and UI state can change between calls. Re-read state when later steps depend on it.

Suite API is intentionally narrow. The coordinator exposes its cross-app workflows; it does not proxy every child server’s tools. Connect an app package separately for direct application tools.

Execution boundaries

Most app tools generate a script for ExtendScript or UXP and send it to a local Adobe app or plugin. Premiere Pro uses its UXP Bridge panel. The tool call is local computer automation with the permissions of the configured process; it is not a hosted service request. Review app-specific notes about unrestricted code-execution tools before using them.