{
  "markdown": "# @picsha-ai/mcp-server\n\nThe official Model Context Protocol (MCP) proxy server for the **Picsha AI** platform.\n\nThis package provides a secure, local `stdio` proxy that connects your LLM and AI agents (like Claude Desktop or OpenClaw) directly to your Picsha AI environment. By running locally, the server is natively enabled to securely read local files and utilize Picsha's direct-to-S3 upload pipelines.\n\n## Installation & Configuration\n\nYou do not need to install this package permanently. You can run it dynamically via `npx`. \n\n### Claude Desktop / OpenClaw\n\nAdd the following to your `claude_desktop_config.json` or `openclaw.json`:\n\n```json\n{\n  \"mcpServers\": {\n    \"picsha-ai\": {\n      \"command\": \"npx\",\n      \"args\": [\n        \"-y\",\n        \"@picsha-ai/mcp-server@latest\"\n      ],\n      \"env\": {\n        \"PICSHA_API_TOKEN\": \"<YOUR_API_TOKEN_HERE>\"\n      }\n    }\n  }\n}\n```\n\n## Security & Multi-Tenancy\n\nYou can generate a `PICSHA_API_TOKEN` via your Picsha Admin Dashboard. By default, this token grants the AI agent administrative access across your entire organization's library.\n\n**Sandbox Mode (User Isolation)**: If you are embedding this MCP server for end-user Slack bots or customer facing SaaS products, you can dynamically restrict the agent's context to a specific user by injecting their User ID as an environment variable:\n\n```json\n      \"env\": {\n        \"PICSHA_API_TOKEN\": \"<YOUR_API_TOKEN>\",\n        \"PICSHA_EXTERNAL_USER_ID\": \"user_123\"\n      }\n```\n\n## Available Tools\n\nThis MCP server provides the following capabilities to your LLM:\n* `search_assets`: Perform exact or hybrid semantic vector searches across your media.\n* `get_asset`: Retrieve deep metadata profiles for resources.\n* `upload_asset`: Automatically infers MIME types and effortlessly uploads local files into Picsha's asynchronous AI ingest pipeline.\n* `generate_render_url`: Provides on-the-fly image transformations and AI generative fill parameters.\n* `trigger_url_ingest`: Ingest public web media directly into the DAM.\n* `moderate_asset`, `link_assets`, `create_dam_group`, `update_asset`, `delete_asset` ...and more!\n\n## Troubleshooting\n\n### Claude Desktop Hangs or Fails to Connect\nIf you are using macOS and Claude Desktop gets stuck connecting to the MCP (or the tools never show up), it is likely due to `npx` dropping standard input/output streams. To fix this:\n\n1. Install the server globally instead of using `npx`:\n   ```bash\n   npm install -g @picsha-ai/mcp-server\n   ```\n2. Update your `claude_desktop_config.json` to point directly to the installed binary:\n   ```json\n   {\n     \"mcpServers\": {\n       \"picsha-ai\": {\n         \"command\": \"picsha-ai-mcp\",\n         \"args\": [],\n         \"env\": {\n           \"PICSHA_API_TOKEN\": \"<YOUR_API_TOKEN_HERE>\"\n         }\n       }\n     }\n   }\n   ```\n3. Restart Claude Desktop.",
  "bytes": 2822,
  "sha": "18bc4ccbf3559aa01451a80cad8d1c94fc7817bac83f70ffaba46d05daca822c",
  "repo_slug": "",
  "fonte": "npm",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_ai_picsha_picsha_9b7697e6/readme"
}