Memory MCP Server

The thin memory MCP server (ainative-zerodb-memory-mcp) provides 18 tools for agent memory, structured reasoning, and write-back actions. 92% smaller than the full server.

npm i ainative-zerodb-memory-mcp

Tools Reference

zerodb_store_memory

Store a memory with content and metadata. Auto-generates embeddings.

Parameters:

Name	Type	Required	Description
content	string	yes	Memory text to store
metadata	object	no	Arbitrary metadata (user_id, source, tags)
importance	number	no	0.0-1.0, default 0.5

Example:

{
  "content": "User prefers dark mode and uses Python",
  "metadata": {"user_id": "u_123", "source": "chat"},
  "importance": 0.8
}

---

zerodb_search_memory

Search memories by keyword or metadata filters.

Parameters:

Name	Type	Required	Description
query	string	yes	Search query
limit	number	no	Max results (default 10)
metadata_filter	object	no	Filter by metadata fields

---

zerodb_semantic_search

Semantic similarity search — finds memories by meaning, not keywords.

Parameters:

Name	Type	Required	Description
query	string	yes	Natural language query
limit	number	no	Max results (default 10)
min_score	number	no	Minimum similarity threshold (0.0-1.0)

---

zerodb_get_context

Get the current session context window — recent memories and key facts for the active session.

Parameters:

Name	Type	Required	Description
session_id	string	no	Session identifier
limit	number	no	Max context items (default 20)

---

zerodb_embed_text

Generate embeddings for text without storing. Useful for pre-processing or comparison.

Parameters:

Name	Type	Required	Description
text	string	yes	Text to embed

Returns: 768-dimensional vector (BAAI/bge-base-en-v1.5).

---

zerodb_clear_session

Clear all session-scoped memories. Does not affect persistent memories.

Parameters:

Name	Type	Required	Description
session_id	string	no	Session to clear (defaults to current)

---

Plan Artifact Tools

Plan artifacts are structured, versioned documents — PRDs, plans, task lists, reasoning traces — stored persistently in ZeroDB with full diff history on every update. Added in v1.2.0.

zerodb_plan_create

Create a new plan artifact.

Parameters:

Name	Type	Required	Description
title	string	yes	Plan title
content	string	yes	Plan body (markdown supported)
tags	array	no	String tags for filtering
metadata	object	no	Arbitrary key-value metadata

Returns: { id, title, content, created_at, version }

Example:

{
  "title": "Q3 API Refactor Plan",
  "content": "## Goals\n- Migrate to async handlers\n- Add rate limiting",
  "tags": ["backend", "q3"]
}

---

zerodb_plan_get

Retrieve the latest version of a plan artifact by ID.

Parameters:

Name	Type	Required	Description
artifact_id	string	yes	Plan artifact ID

Returns: Latest version of the plan with full content and metadata.

---

zerodb_plan_update

Update a plan artifact. Every update creates a new version entry — full diff history is preserved.

Parameters:

Name	Type	Required	Description
artifact_id	string	yes	Plan artifact ID
content	string	yes	New plan content
title	string	no	Updated title
tags	array	no	Updated tags

Returns: Updated plan with new updated_at and version.

---

zerodb_plan_history

Get the full version history of a plan artifact, ordered newest first.

Parameters:

Name	Type	Required	Description
artifact_id	string	yes	Plan artifact ID

Returns: Array of all versions with timestamps and content diffs.

---

Auto-Context Tools

Auto-context automatically injects relevant memories into agent prompts without explicit recall calls.

zerodb_configure_auto_context

Enable or disable automatic context injection for the current session.

Parameters:

Name	Type	Required	Description
enabled	boolean	yes	Enable/disable auto-context
max_memories	number	no	Max memories to inject (default 5)
min_relevance	number	no	Minimum similarity threshold (default 0.6)
auto_trace	boolean	no	Auto-record decision traces (default false)

Example:

{
  "enabled": true,
  "max_memories": 8,
  "min_relevance": 0.7,
  "auto_trace": true
}

When enabled, the MCP server automatically:

1. Embeds the user's latest message
2. Searches for relevant memories above min_relevance
3. Injects them as context before the agent responds
4. Optionally records decision traces for each interaction

note

Auto-context is independent of consolidation. Consolidation runs on schedule regardless of whether auto-context is enabled.

---

zerodb_get_auto_context_config

Get the current auto-context configuration for the active session.

Parameters: None

Returns:

{
  "enabled": true,
  "max_memories": 5,
  "min_relevance": 0.6,
  "auto_trace": false
}

---

zerodb_synthesize_context

Synthesize a natural-language context summary from relevant memories. Unlike get_context which returns raw memories, this uses an LLM to produce a coherent narrative.

Parameters:

Name	Type	Required	Description
query	string	yes	What context to synthesize about
max_sources	number	no	Max source memories to consider (default 10)

Returns:

{
  "synthesis": "The user is a senior backend engineer who prefers...",
  "sources": ["mem_abc", "mem_def"],
  "confidence": 0.85
}

---

Write-Back Action Tools

These tools allow agents to take actions based on memory insights — triggered by consolidation, auto-context, or agent decision.

zerodb_slack_send

Send a Slack message to a connected channel.

Name	Type	Required	Description
channel	string	yes	Slack channel name or ID
message	string	yes	Message content

zerodb_gmail_reply

Reply to a Gmail thread.

Name	Type	Required	Description
thread_id	string	yes	Gmail thread ID
body	string	yes	Reply body

zerodb_calendar_create

Create a calendar event.

Name	Type	Required	Description
title	string	yes	Event title
start	string	yes	ISO 8601 start time
end	string	yes	ISO 8601 end time

zerodb_github_create_issue

Create a GitHub issue.

Name	Type	Required	Description
repo	string	yes	Repository (owner/name)
title	string	yes	Issue title
body	string	no	Issue body

zerodb_notion_create_page

Create a Notion page.

Name	Type	Required	Description
parent_id	string	yes	Parent page or database ID
title	string	yes	Page title
content	string	no	Page content (markdown)

---

Configuration

{
  "mcpServers": {
    "zerodb-memory": {
      "command": "npx",
      "args": ["-y", "ainative-zerodb-memory-mcp"],
      "env": {
        "ZERODB_API_KEY": "your-api-key",
        "ZERODB_PROJECT_ID": "your-project-id"
      }
    }
  }
}

When to Use This vs Full Server

Use Case	Memory Server	Full Server
Agent memory (store/recall)	Yes	Yes
Plan artifacts (versioned docs)	Yes	Yes
Auto-context injection	Yes	Yes
Context synthesis (LLM)	Yes	Yes
Write-back actions (Slack, Gmail, etc.)	Yes	No
Vector search	No	Yes
File storage	No	Yes
NoSQL tables	No	Yes
PostgreSQL	No	Yes
Bundle size	~50KB	~500KB

Use the memory server when your agent only needs memory and structured plans. Use the full server when you need the complete ZeroDB data layer.

Related

• Sequential Thinking MCP — persistent reasoning chains built on plan artifacts
• ZeroMemory Overview — cognitive memory architecture