@astramindapp/mcp-server

MIND MCP Server — The most complete AI memory layer available. 44 tools, 384 actions.

Your AI agents forget everything between sessions. MIND fixes that. Connect any MCP-compatible agent to your personal knowledge graph — with emotional intelligence, CRM, life management, site-wide assignable tasks, social features, self-training, autonomous insights, an admin-only Agent Command Center with per-agent ticket queues, an admin-only Featured Minds Portal (curate the public marketplace at m-i-n-d.ai from any MCP client — model, public chat prompt, temperature, brand fields, write-through to the linked user's profile), and more.

Self-describing. Every connecting client receives a full integration briefing in the MCP initialize response, and can read the complete playbook from the mind://integration-guide resource — so any agent knows what MIND is, which integration path fits its runtime (MCP server / OpenClaw plugin / REST API), how to authenticate, and the session protocol, without being told.

44 Tools

Quick Start

Fastest: one-paste remote connect (Claude Code / Claude Desktop / claude.ai)

No install, no API key handling — paste one command, click Connect in the browser, done:

claude mcp add --transport http mind https://www.m-i-n-d.ai/mcp

Claude opens the m-i-n-d.ai consent page, you click Connect, and MIND is live. Auth, key minting, and refresh are all handled by OAuth.

No browser OAuth in your host? Use a key on the same URL

Every host that lets you add a custom MCP server with a header can use MIND with no OAuth and no npm install: same remote server URL, just swap in a key.

• Server URL: https://www.m-i-n-d.ai/mcp
• Header: Authorization: Bearer mind_<your key>
• Mint a key: https://www.m-i-n-d.ai → Settings → Developer → API Keys

Agents: if AuthenticateMcpServer or the host's connect card fails, ask the user for a MIND API key and rewire the same URL with that header — do not fall back to the local npm server just because OAuth didn't work.

Local stdio install: browser-connect setup wizard

For tools that run MCP servers locally (Claude Desktop, Cursor, Windsurf, VS Code Copilot), run the setup wizard — it opens your browser, you click Connect on m-i-n-d.ai, and it receives your API key automatically and writes every tool config for you:

npx -y --package=@astramindapp/mcp-server mind-mcp-setup

No manual key copying. The wizard auto-detects your installed AI tools and configures each one.

Manual alternative: paste an API key

If you'd rather mint a key yourself: sign up at www.m-i-n-d.ai → Settings → Developer → Create API Key, then:

npx -y --package=@astramindapp/mcp-server mind-mcp-setup --key mind_xxx

or install globally and configure by hand:

npm install -g @astramindapp/mcp-server

Claude Code (local stdio with a manual key)

claude mcp add mind -- env MIND_API_KEY=mind_xxx mind-mcp

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "mind": {
      "command": "mind-mcp",
      "env": {
        "MIND_API_KEY": "mind_your_key_here"
      }
    }
  }
}

The config file lives at ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, or %APPDATA%\Claude\claude_desktop_config.json on Windows. Fully quit and reopen Claude Desktop after editing — it only reads this file on launch.

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "mind": {
      "command": "mind-mcp",
      "env": {
        "MIND_API_KEY": "mind_your_key_here"
      }
    }
  }
}

Any MCP-Compatible Agent

The server works with any tool that supports Model Context Protocol — Claude, GPT, Gemini, Llama, or any future model.

How It Works

Your AI Agent  <-->  MCP Protocol  <-->  MIND MCP Server  <-->  Personal Knowledge Graph
                                                                  |
                                                           Emotional Intelligence
                                                           Autonomous Learning
                                                           CRM + Life + Social

1. Agent calls mind_context at session start → loads identity, rules, priorities
2. Agent calls mind_query before decisions → retrieves relevant memories
3. Agent calls mind_remember after tasks → stores outcomes and learnings
4. Agent calls mind_sense → reads user's emotional state to adapt responses
5. Agent calls mind_insights → surfaces autonomous pattern detection
6. Next session, any agent has full context. Knowledge compounds.

What Makes MIND Different

vs. Flat File Memory (MEMORY.md)

vs. Other MCP Memory Servers

Tool Reference

mind_query — Search Your Knowledge Graph

query: "What did I decide about the authentication approach?"
mode: "hybrid"  // hybrid (default), mix, global, local, naive

Returns RETRIEVED CONTEXT from your stored documents, entries, and feed posts with source attribution — raw graph context for the calling model to synthesize, not a finished answer. Uses no MIND LLM and costs 0 credits. Pass retrieve_only=false for a MIND-written answer (spends credits).

mind_remember — Store & Manage Content

mind_folders — Organize Documents into Folders

Folders are a presentation layer over the document tray — the knowledge graph still indexes and retrieves across every document regardless of folder.

mind_context — Load Persistent Context

Loads five structured sections at session start:

• Soul — Core identity, mission, personality
• User — Who the user is, their role, preferences
• Rules — Operating constraints, behavioral guidelines
• Priorities — Current goals, active projects, deadlines
• Recent — Latest activity, outcomes, decisions

mind_life — Life Management + Calendar

mind_crm — Contact Relationship Management

mind_tasks — Site-Wide Tasks

Assignable, completable, reportable work items. A task can attach to a Life project (parent_type: "life_item"), a CRM contact ("contact"), an agent ("agent"), or stand alone. Assign tasks to a MIND member, an agent, or an external email address.

mind_tasks({ action: "create", title: "Follow up with Acme",
             parent_type: "contact", parent_id: "<contact_id>",
             assignee_type: "agent", assignee_id: "fundraising-agent",
             dispatch_agent: true, due_date: "2026-06-01" })

mind_tasks({ action: "reports" })

mind_sense — MINDsense Emotional Intelligence

mind_research — Autonomous Research

mind_train — Self-Training

mind_social — Social Layer

mind_profile — Profile & Preferences

mind_insights — Autonomous Learning Engine

mind_automate — Automations

mind_notify — Notifications

mind_admin — Administration & Featured Minds Portal

Requires an admin-scoped MIND API key. Powers everything at https://m-i-n-d.ai/#/admin/featuredmindsportal — the same surface any MCP client (Claude Code, Cursor, Windsurf, n8n via MCP, custom agents) can drive end-to-end.

User provisioning + tier/credits

Featured Minds Portal — full CRUD

The catalog (featured_minds collection) and the linked owner profile (user_profiles) edited together so changes hit https://m-i-n-d.ai/m/{username} immediately.

Recipe — change a featured MIND's model + prompt from MCP

mind_admin({ action: "list_featured_minds" })
// → grab the mind_id you want

mind_admin({ action: "get_featured_mind_full", mind_id: "<id>" })
// → see current owner_profile + the full available_models catalog

mind_admin({
  action: "update_featured_mind_owner_profile",
  mind_id: "<id>",
  preferred_llm_model: "anthropic/claude-sonnet-4.6",
  chat_temperature: 0.7,
  public_mind_prompt: "You are Anthony's public MIND. Answer in his voice."
})
// → next anonymous chat at /m/{username} uses the new model + prompt + temp

mind_agents — Agent Command Center (Admin)

Canonical registry of every agent across the workspace fleet — running on a VPS, planned, archived. Solves the recurring "every Claude Code session re-discovers my agents from scratch" problem. Backed by /admin/agents on the MIND backend. Requires an admin API key.

// Before scaffolding a new agent — check what exists
mind_agents({ action: "list" })

// After non-trivial work — keep the registry honest
mind_agents({
  action: "heartbeat",
  slug: "social-media-manager",
  current_job: "drafting LinkedIn carousel for Friday"
})

// Promote a planned agent to running
mind_agents({ action: "set_status", slug: "audit-agent", status: "running" })

// Hand an agent to a teammate's board, or share it read-only
mind_agents({ action: "transfer_owner", slug: "audit-agent", owner_username: "jane" })
mind_agents({ action: "share", slug: "audit-agent", grantee_username: "sam", share_role: "viewer" })

// Search the fleet by free text — slug, name, description, or tags
mind_agents({ action: "list", list_query: "atlas" })

UI: visit https://m-i-n-d.ai/agents (admin-only) to browse the fleet visually.

mind_tickets — Agent Ticket Queue (Admin)

Every agent in the Command Center carries a ticket queue — client feedback, critique, ideas, feature requests, and bugs. A consulting client an agent is shared with files tickets; the agent's owner triages and resolves them. Every ticket auto-assigns to the Ernie triage agent and mirrors into MIND Life as a task. Backed by /admin/agents/{slug}/tickets. Requires an admin API key. agent_slug is required on every call.

kind: feedback · critique · idea · feature · bug status: open → triaged → in_progress → resolved → closed priority: low · medium · high · urgent

// File a bug against an agent
mind_tickets({ action: "create", agent_slug: "social-media-manager",
               kind: "bug", priority: "high",
               title: "Carousel export drops the last slide",
               body: "Repro: 7-slide deck exports as 6." })

// Answer it, then resolve it
mind_tickets({ action: "comment", agent_slug: "social-media-manager",
               ticket_id: "<id>", body: "Fixed in the export batch size — verified." })
mind_tickets({ action: "resolve", agent_slug: "social-media-manager", ticket_id: "<id>" })

mind_accounts — Multi-MIND Accounts

A "MIND" is a knowledge-graph account. One person can own — or be granted access to — many MINDs (work, research, a client's, a shared team MIND) and switch between them. This tool lets an agent discover and manage them. Backed by /developer/v1/accounts on the MIND backend.

// Discover every MIND you can reach
mind_accounts({ action: "list" })

// Create a dedicated MIND for a new workstream
mind_accounts({ action: "create", label: "Acme Corp Research" })

// Grant a teammate read-only access
mind_accounts({
  action: "grant",
  mind_username: "acme_research_a1b2c3",
  grantee_username: "teammate",
  role: "viewer"
})

Note: owner = full access, viewer = read-only. Requires MULTI_MIND_ACCOUNTS_ENABLED on the MIND backend; list returns enabled: false when the feature is off.

Partner Integration

Partner apps can programmatically create MIND accounts using partner keys:

curl -X POST https://m-i-n-d.ai/admin/users/create \
  -H "X-API-Key: mind_partner_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "username": "newuser",
    "email": "user@app.com",
    "password": "securepassword",
    "generate_api_key": true
  }'

Returns a JWT + permanent API key. Your app stores the API key and uses it for all subsequent MCP/API calls on behalf of that user.

Environment Variables

Programmatic Usage

import { createMindMcpServer, MindClient } from "@astramindapp/mcp-server";

const client = new MindClient({
  baseUrl: "https://www.m-i-n-d.ai",
  apiKey: "mind_xxx",
});

const server = createMindMcpServer(client);
// Connect to any MCP transport

Patents

MIND's core technology — including emotion-weighted knowledge graph encoding and cross-agent persistent memory via the Model Context Protocol — is the subject of pending patent applications. Patents pending.

Troubleshooting

• "Connect" opened the MIND website instead of connecting — this was a bug in the non-www endpoint, fixed July 2026. Use https://www.m-i-n-d.ai/mcp (with the www) and retry.
• No "Connectors" option in my tool — use the Claude Code one-liner above if you're on Claude Code, or the local npm install (npm install -g @astramindapp/mcp-server) for any other MCP-compatible tool.
• npx says "could not determine executable to run" / hangs — npx -y <pkg> only runs the bin named after the package's last path segment (mcp-server), so npx -y @astramindapp/mcp-server alone fails, and adding @latest plus a trailing mind-mcp-setup just passes it as an argument instead of selecting that bin — use npx -y --package=@astramindapp/mcp-server mind-mcp-setup (or mind-mcp), which works on every version, or install globally first (npm install -g @astramindapp/mcp-server).
• I have multiple MIND accounts — add the connector once per account under a different name (e.g. mind-personal, mind-work); each Connect click lets you pick the account on the consent screen.

Links

• Website: www.m-i-n-d.ai
• GitHub: theastraway/mind-mcp-server
• npm: @astramindapp/mcp-server
• MCP Docs: m-i-n-d.ai/mcp.html

License

MIT — Astra AI, Inc.