@multiplist/mcp-server

Portable AI Memory Vault & Workspace
Connect Claude Desktop, Claude Code, Cursor, Antigravity IDE, and frontier LLMs directly to your private Multiplist brain.

🌟 What is Multiplist?

Your AI conversations are high-density intellectual property. Most memory tools save flat markdown text or generic summaries. Multiplist treats your thinking as structured capital:

1. 8 Canonical Categories of Meaning: Automatically mines conversations for decisions, frameworks, golden passages, definitions, actions, questions, offers, and emergence.

2. Exact Provenance: Every claim, decision, and model traces back to exact line and character boundaries in your source conversations ([Seed <uuid>]).

3. Continuous Workspaces: Tactile Bento containers, studio desks, 5-lane progression boards (📥 Intake ➔ ⚡ Doing ➔ ⏸️ Needs Context ➔ 👁️ Review ➔ 🚢 Shipped), and card graduation.

4. The Shared Office for Human and AI: Continuous workspace coordination (Bento studios, decks, and cards) that connects chat decisions to real-world project tracking.

5. Zero Data Custody: Transcripts and extractions write directly into your private Multiplist vault.

Related MCP server: Memory Nexus

🚀 Quickstart

1. Claude Desktop (stdio Proxy)

Add Multiplist to your claude_desktop_config.json:

{
  "mcpServers": {
    "multiplist": {
      "command": "npx",
      "args": ["-y", "@multiplist/mcp-server@latest"],
      "env": {
        "MULTIPLIST_MCP_TOKEN": "mcp_live_YOUR_TOKEN_HERE"
      }
    }
  }
}

🔑 Get your MCP Token: Log in to multiplist.ai ➔ Settings ➔ MCP Tokens ➔ Generate a live token (prefix mcp_live_).

2. Direct Hosted StreamableHTTP

For remote agents or clients supporting StreamableHTTP:

{
  "mcpServers": {
    "multiplist": {
      "url": "https://multiplist.ai/mcp",
      "headers": {
        "x-mcp-key": "mcp_live_YOUR_TOKEN_HERE"
      }
    }
  }
}

🛠️ 71 Canonical Tools Across 4 Tiers

Multiplist features progressive tool disclosure matching your provisioned workspace tier:

🟢 Core Tools (12 Free / BYOK)

• create_source: Save a conversation, document, or transcript to the vault with automatic extraction.

• update_source: Append or update saved conversations with incremental extraction.

• delete_source: Soft-delete a source while preserving extracted seed provenance.

• trigger_extraction: Run single-source extraction across the 8 canonical categories.

• get_source: Retrieve a complete source with its abstract, seeds, and character citations.

• search_vault: Semantic recall for extracted seeds across all conversations.

• search_sources: Browse the catalog of sources by title, date, or keyword.

• get_vault_summary: Vault-wide overview of sources, seed counts, and category clusters.

• get_vault_gaps: Surface unextracted areas, sparse categories, and missing knowledge.

• get_capabilities: Session orientation and progressive disclosure introspection.

• list_local_models: Discover locally available Ollama models at $0 compute.

• query_local_model: Execute zero-cost local LLM inference against local models.

🔵 Plus Tools (48 Workspace Tools)

• Spatial Workstations: compose_workspace, resolve_workspace, list_containers, create_container, update_container, delete_container, get_container, get_container_structure, get_container_stats

• Studios & Pins: list_studios, create_studio, update_studio, delete_studio, get_deck_state, pin_source, unpin_source, list_pins

• Decks, Lanes & Cards: list_lanes, create_lane, update_lane, delete_lane, get_lane_contents, list_cards, get_card, create_card, update_card, move_card, delete_card, graduate_card

• Research & Folders: request_brief, get_brief_status, create_vault_folder, delete_vault_folder, rename_vault_folder, add_sources_to_folder, remove_source_from_folder, batch_extract, batch_status

• Custom Domain Skills: create_skill, list_skills, get_skill, match_skills, update_skill, delete_skill

• Living Marginalia & Assets: add_annotation, add_marginalia, list_source_assets, rescue_excerpt

🟣 Pro Tools (6 Autonomy Tools)

• create_default_studio_lanes: Create standard workflow lanes (Intake, Doing, Needs Context, Review, Shipped) for a studio workspace. Idempotent — will not duplicate existing lanes.

• get_desk_state: Inspect active desk cards, agent configs, and tool bouquets.

• query_agent_log: Audit autonomous agent execution history and tool invocations.

• list_production_skills: Discover Assemblist production deliverable templates.

• produce_deliverable: Generate holographic capsules, executive briefs, and deliverables.

• get_context: Fetch full context capsule snapshots for high-reasoning tasks.

🌐 Realm Tools (5 Mesh Tools)

• list_realms, get_current_realm, switch_realm, share_source_to_realm, move_source_to_realm: Sovereign domain isolation and zero-token cross-vault asset sharing.

💬 18 Interactive MCP Prompts

Multiplist exports 18 interactive prompt templates with typed arguments for 1-click execution in Claude Desktop, Cursor, and any MCP host:

🧬 Provenance & 8 Canonical Categories

When you save content, Multiplist extracts 8 canonical categories:

• Decisions: Commitments made, positions locked, rationale cited.

• Frameworks: Mental models, named structures, repeatable systems.

• Golden Passages: Key quotes and phrasing worth preserving verbatim.

• Definitions: Terminology and domain definitions established in context.

• Actions: Concrete deliverables and commitments with owners.

• Questions: Unresolved inquiry vectors and deliberate open loops.

• Offers: High-leverage candidate actions and propositions.

• Emergence: Nascent concepts forming in dialogue.

📜 Public Discovery & Directory Specifications

• Hosted Server Card: https://multiplist.ai/.well-known/mcp/server-card.json

• Glama Manifest: glama.json (root)

• MCP Registry Manifest: server.json (root)

• Smithery Registry: @multiplist/mcp-server

📄 License

MIT © Multiplist & Mystic Quarterly

Session cleanup

On stdin EOF, SIGINT, or SIGTERM, the bridge stops accepting messages, waits for already dispatched requests (including initialization), and requests authenticated session termination before closing its transports. Repeated shutdown events share one shutdown operation. Draining and termination have a combined 3-second deadline; local transport closure has a further 1-second deadline.

A failed or timed-out termination produces a short stderr warning and the process still exits. Capacity is only released if the host accepts termination. Unsupported DELETE responses, lost responses, forced process kills, and network failures can leave a session for the host's existing retention/reaping policy. The bridge does not change that policy or the host's security limits.

Run npm test for the build and loopback-only shutdown/capacity regression suite.