npx agentspec-cli install 8927cfd1-225d-4d15-9bb1-626b161e353f --file-type agents-md→ .agents/
# AGENTS.md — dcc-mcp-photoshop
> **Navigation map, not a reference manual.**
> Follow the links; don't read everything upfront.
> Agent-specific files (`CLAUDE.md`, `GEMINI.md`, `COPILOT.md`) intentionally point back here.
## Agent Control Path
AI agent runtimes default to the shared gateway through the
`dcc-mcp` skill and `dcc-mcp-cli` REST commands:
```bash
dcc-mcp-cli search --query "<task>" --dcc-type photoshop
dcc-mcp-cli describe <tool-slug>
dcc-mcp-cli call <tool-slug> --json '{"key":"value"}'
```
Use `dcc-mcp-cli list` for live instances and `dcc-mcp-cli dcc-types` for
release-catalog support. IDE users may continue to configure the gateway MCP
endpoint; adapter-local Python start APIs are for host bootstrap and tests.
### CLI availability and updates
If `dcc-mcp-cli` is missing, obtain user consent before using the official
install commands in the README Agent workflow. Keep an official build current
with:
```bash
dcc-mcp-cli update check
dcc-mcp-cli update apply
```
`update apply` stages the latest CLI for the next launch; it does not replace
a running server.
## Project Overview
**dcc-mcp-photoshop** turns Adobe Photoshop into an MCP Streamable HTTP backend via a UXP WebSocket plugin. The Python bridge runs a WebSocket server on port 9001; the UXP plugin inside Photoshop connects as a client.
## Agent Entry Path
```
AI Agent
│ MCP Streamable HTTP → http://127.0.0.1:9765/mcp
▼
dcc-mcp-server Gateway (auto-discovers DCC via capability index)
│
▼
PhotoshopMcpServer [Python sidecar]
│ WebSocket JSON-RPC (port 9001)
▼
UXP Plugin [JavaScript, runs inside Photoshop]
│ UXP API calls
▼
Adobe Photoshop 2022+
```
## Entry Strategy
| Scenario | Path |
|----------|------|
| AI agent / CLI runtime | `dcc-mcp` + `dcc-mcp-cli` over gateway REST `/v1/search`, `/v1/describe`, and `/v1/call` |
| IDE user (Cursor, Claude Desktop) | Configure `mcpServers` → `url: "http://127.0.0.1:9765/mcp"` |
| Development / debugging | Embedded mode: `dcc-mcp-photoshop --embedded` (MCP server + bridge in one process) |
## Skills-First Workflow
When using Photoshop through dcc-mcp-photoshop, prefer skills over raw scripting:
```
1. SEARCH: search_skills(query="photoshop") → find available skill packages
2. CHECK: Read the skill's SKILL.md description and tools
3. LOAD: load_skill("photoshop-document") → expose the tools
4. CALL: Call the specific tool with validated parameters
5. FOLLOW UP: Check structured results for next steps
```
### Available Skills
| Skill | Tools | Purpose |
|-------|-------|---------|
| `photoshop-setup` | 6 | Install, configure, verify bridge connection |
| `photoshop-document` | 2 | Document info, list layers |
| `photoshop-image` | 7 | Create document, export, resize, flatten, merge |
| `photoshop-layers` | 8 | Layer CRUD, opacity, visibility, blend mode, fill |
| `photoshop-text` | 3 | Create, update, inspect text layers |
## CLI Modes
| Mode | Command | Use Case |
|------|---------|----------|
| Bridge-only (default) | `dcc-mcp-photoshop` | Deployment with external `dcc-mcp-server` |
| Embedded | `dcc-mcp-photoshop --embedded` | Development (MCP + bridge in one process) |
| Daemon | `dcc-mcp-photoshop --daemon` | Non-interactive background startup |
## Distribution Channels
| Channel | Artifact | Python Required |
|---------|----------|-----------------|
| PyPI | `dcc-mcp-photoshop` wheel + sdist | Yes (3.8+) |
| GitHub Release | Standalone binary (Win/Linux/Mac) | No |
| GitHub Release | UXP `.ccx` plugin | No |
## Key Files
| File | Purpose |
|------|---------|
| `README.md` | Human-readable project documentation (EN) |
| `README_zh.md` | Human-readable project documentation (ZH) |
| `CHANGELOG.md` | Release history |
| `docs/PRD.md` | Product Requirements Document |
| `docs/bridge-protocol.md` | WebSocket JSON-RPC 2.0 bridge protocol spec |
| `docs/distribution.md` | Distribution & release guide |
| `src/dcc_mcp_photoshop/` | Python package source |
| `bridge/uxp-plugin/` | UXP plugin (JavaScript, runs inside Photoshop) |
## Response Language
- Reply to the user in **Simplified Chinese** by default.
- Keep all code, identifiers, commit messages, and file contents in **English**.
## PR / Commit Rules
- No AI-attribution footers in PR bodies or commit messages.
- Rebase onto main before merging — no merge commits.
- CI must pass before review.
Multi-Agent Gemini MCP Server Setup Guide
This guide will walk you through setting up a multi-agent MCP server that provides specialized AI agents (planner, reviewer, architect) to work with Claude Code.
by AgentHub Crawler
AGENTS.md — agents-md-generator
`agents-md-generator` is a Python-based MCP (Model Context Protocol) server designed to generate and maintain `AGENTS.md` files—architectural context documents for AI coding agents. It uses Tree-sitter for high-fidelity AST analysis across multiple languages (Python, TypeScript, Go, C#) and `uv` for dependency management. The system operates as an incremental, cache-aware pipeline that transforms a codebase into a structured payload for AI assistants to perform maintenance tasks without manual e
by AgentHub Crawler
AGENTS.md — Coding Agent Guidelines
This file provides guidance for agentic coding agents working in this repository.
by AgentHub Crawler
Sign in to leave a comment.
AGENTS.md — AI Agent Guidelines for fVDB Reality Capture
This file provides persistent instructions for AI coding agents (Cursor, Copilot, Codex, etc.) working on this codebase. Human contributors should also refer to [CONTRIBUTING.md](CONTRIBUTING.md).
by AgentHub Crawler