DocsMCP Server

MCP Server

Your identity in every AI tool

Connect Artificial ID to Claude Desktop, Cursor, Cline, Windsurf, and any other MCP-compatible client with a single config change. Real-time blueprint sync, no manual copy-paste.

Install command
npx @artificial-id/mcp

On this page

What is MCPSupported clientsConfigurationToolsPromptsSecurity modelTroubleshooting

Overview

What is the Model Context Protocol?

The Model Context Protocol (MCP) is an open standard created by Anthropic that lets AI clients (like Claude Desktop or Cursor) connect to external servers that provide tools, prompts, and resources. Instead of copy-pasting your identity into every chat, you connect the Artificial ID MCP server once and every supported AI client has live access to your blueprint.

One-time setup

Add 8 lines of JSON to your AI client config. The server runs locally via npx — no installation, no Docker, no background services.

Always in sync

Your blueprint is fetched fresh on every request (with a 60s cache). Update a module in the dashboard and your next AI conversation sees it immediately.

Scoped access

MCP keys can be scoped to read-only, preventing the server from writing outputs or signing watermarks. Give your AI tools only the permissions they need.

Supported clients

Works with your tools

Any MCP-compatible client works. These four are tested and verified.

Claude Desktop

Supported

Supports all 5 tools and the system_context prompt. Blueprint auto-reloads on each conversation.

~/Library/Application Support/Claude/claude_desktop_config.json

Cursor

Supported

Full tool support. Use get_profile in Cursor Rules to inject identity into every AI interaction.

.cursor/mcp.json in your project root

Cline

Supported

Full tool and prompt support. Works with all Cline AI providers.

VS Code settings: mcp.servers

Windsurf

Supported

Full support. Cascade AI will read your blueprint on each agentic task.

~/.codeium/windsurf/mcp_config.json

Configuration

Add to your MCP config

Open your client's MCP configuration file and add the artificial-id server block. Replace aiid_live_your_key_here with an API key generated from your dashboard.

mcp_config.json
json
{
  "mcpServers": {
    "artificial-id": {
      "command": "npx",
      "args": ["-y", "@artificial-id/mcp"],
      "env": {
        "AIID_API_KEY": "aiid_live_your_key_here"
      }
    }
  }
}

No global install needed

The -y flag in the args auto-accepts the npx prompt. The MCP server package is downloaded on first run and cached by npm. You do not need to run any install command manually.

Use a dedicated MCP key

Generate a separate API key for MCP use from your dashboard. This lets you revoke access to your AI tools independently without affecting your main API key or browser extension.

Available tools

5 MCP tools

Tools are callable by the AI model during a conversation. The model decides when to call them based on context — or you can instruct it directly.

get_profileread-only

Returns the full decrypted blueprint for the authenticated user. Includes all 4 layers and all modules. This is the primary tool for giving an AI model full context about who it is talking to.

Example call

// No parameters needed
// Returns the full blueprint JSON
get_personaread-only

Returns a specific persona by ID, with all module values resolved (base blueprint + overrides merged). Use this when you want the AI to respond through a specific context lens.

Parameters

persona_idstringrequiredThe ID of the persona to retrieve.

Example call

{ "persona_id": "ps_work_mode_abc123" }
get_moduleread-only

Retrieves a single module value from the blueprint. Useful for targeted lookups without loading the full blueprint — for example, just fetching preferred code style or communication preferences.

Parameters

module_idstringrequiredDot-path to the module, e.g. code_dna.style

Example call

{ "module_id": "code_dna.languages" }
log_outputread-write

Logs an AI-generated output against the active blueprint. Optionally signs the output with an HMAC-SHA256 watermark for authorship verification.

Parameters

contentstringrequiredThe AI-generated text to log.
persona_idstringActive persona at time of generation.
signbooleanWhether to attach a watermark signature.

Example call

{
  "content": "Here is the refactored function...",
  "persona_id": "ps_code_review",
  "sign": true
}
watermark_contentread-write

Signs a piece of content with an HMAC-SHA256 watermark tied to the user's blueprint. Returns the signature. Use this when you want provable authorship without full output logging.

Parameters

contentstringrequiredThe text to watermark.

Example call

{ "content": "My analysis of the architecture..." }

Available prompts

MCP prompts

Prompts are pre-built system message templates your AI client can load on demand. Unlike tools, prompts are inserted into the conversation context rather than called during inference.

system_context

A pre-formatted system prompt built from the current user's blueprint. Inject this at the start of any conversation to give the AI complete identity context. The prompt is assembled from all 4 layers, optimized for token efficiency.

Parameters

persona_idstring?Optionally scope the context to a specific persona.
layersstring[]?Optionally limit to specific layers (e.g. ["core_identity", "code_dna"]).

Security model

API key scoping

MCP API keys support two permission tiers. You can restrict a key to read-only so that your AI tools can read your blueprint but cannot write outputs or sign watermarks on your behalf.

Read-only scope

Use this for AI coding assistants that only need to know who you are. Recommended for Cursor and Cline.

get_profile
get_persona
get_module
system_context prompt
log_output
watermark_content

Read-write scope

Required for Claude Desktop if you want the AI to log its own outputs and apply watermarks. Generate a dedicated key for this purpose.

get_profile
get_persona
get_module
log_output
watermark_content
system_context prompt

Troubleshooting

Common issues

Server fails to start / npx download error

Ensure Node.js 18+ is installed. Run `npx @artificial-id/mcp --version` in your terminal first to force the package download. If behind a corporate proxy, set the HTTPS_PROXY environment variable.

Authentication error: Invalid or expired API key

Verify your key starts with aiid_live_ and has not been revoked. Check the AIID_API_KEY environment variable is set correctly in the config — it must be inside the "env" object, not at the top level.

Blueprint data is stale / not updating

The MCP server caches your blueprint for 60 seconds to reduce API calls. If you just updated a module, wait 60 seconds and re-run the get_profile tool. You can set AIID_CACHE_TTL=0 in the env to disable caching during development.

Tools show up but return empty data

This usually means your blueprint interview is incomplete. Log into your dashboard and check that the interview is finished and at least one module has data. The get_module tool will return null for empty modules.

Next steps

Related documentation

API Reference

Explore the full REST API — the same endpoints the MCP server calls under the hood.

Read docs

Security & Encryption

How your blueprint stays zero-knowledge even when accessed through the MCP server.

Read docs

Getting Started

Create your account and complete the interview before setting up MCP.

Read docs