Skip to main content
Glama

@kolivri/memoria-mcp — Deterministic AI Memory & Code Intelligence MCP for Claude Code, OpenCode, Cursor & Claude Desktop

kolivritech/memoria-mcp MCP server Node.js >= 18

@kolivri/memoria-mcp is the official zero-dependency Node.js ESM Model Context Protocol (MCP) connector (stdio & HTTP) for Memoria Cloud — giving AI coding and operator agents (Claude Code, OpenCode, Cursor, Windsurf, VS Code, Cline, Roo Code, Zed, Goose, and Claude Desktop) deterministic long-term memory, cross-session recall, and code intelligence powered by memoria.kolivri.com.

  • Memoria Cloud Portal & API Keys: https://memoria.kolivri.com (Get API Key — includes 1,000 free API calls/memories; paid tiers: Developer $19/mo, Startup $79/mo, Business $299/mo)

  • Documentation & Benchmarks: https://memoria.kolivri.com/docs (500/500 = 100.0% LongMemEval at 12.9 ms p50, 102/102 = 100.0% Hebrew LongMemEval across 15 languages)

  • Support: stas@kolivri.com · WhatsApp 054-787-5670


Environment Variables

Variable

Required

Default

Description

MEMORIA_API_KEY

Yes

—

Your Memoria Cloud API key (mem_dev_..., mem_start_..., or mem_biz_...) from memoria.kolivri.com/register

MEMORIA_API_URL

Optional

https://memoria.kolivri.com

Memoria Cloud base URL

MEMORIA_NAMESPACE

Optional

default

Default project/workspace namespace when omitted from tool arguments


Related MCP server: io.github.cdnsteve/sugar

Installation

1. One-Line Auto-Installer (15 AI Clients + Claude Code Lifecycle Hooks)

Automatically configures Claude Code, Cursor, Windsurf, VS Code, Claude Desktop, OpenCode, Cline, Roo Code, Continue, Zed, Goose, PearAI, Trae, Void, and Melty:

# macOS / Linux / WSL
curl -fsSL --proto '=https' --tlsv1.2 https://memoria.kolivri.com/downloads/install.sh | MEMORIA_KEY=YOUR_MEMORIA_API_KEY bash
# Windows PowerShell
$env:MEMORIA_KEY="YOUR_MEMORIA_API_KEY"; irm https://memoria.kolivri.com/downloads/install.ps1 | iex

2. Claude Code (claude mcp add)

claude mcp add memoria \
  -e MEMORIA_API_KEY="your-memoria-api-key" \
  -- npx -y @kolivri/memoria-mcp

3. Cursor (~/.cursor/mcp.json) & Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "memoria": {
      "command": "npx",
      "args": ["-y", "@kolivri/memoria-mcp"],
      "env": {
        "MEMORIA_API_KEY": "your-memoria-api-key",
        "MEMORIA_NAMESPACE": "my-project"
      }
    }
  }
}

4. OpenCode (~/.config/opencode/opencode.json)

{
  "mcp": {
    "memoria": {
      "type": "local",
      "command": ["npx", "-y", "@kolivri/memoria-mcp"],
      "environment": {
        "MEMORIA_API_KEY": "your-memoria-api-key",
        "MEMORIA_NAMESPACE": "my-project"
      },
      "enabled": true
    }
  }
}

5. Docker

docker build -t memoria-mcp .
docker run -i --rm \
  -e MEMORIA_API_KEY="your-memoria-api-key" \
  -e MEMORIA_NAMESPACE="my-project" \
  memoria-mcp

14 MCP Tools (https://memoria.kolivri.com/v1/mcp)

Tool

Required Args

Description

encode_memory

content

Store a new memory in Memoria Cloud with tier (episodic, semantic, buffer, core), salience (0.0–1.0), namespace, and context pairs.

batch_encode

memories

Atomically encode an array of memories in a single call.

recall_memories

query

Retrieve top memories ranked deterministically by Memoria Cloud within token_budget.

build_context

query

Assemble a token-budgeted <memory_context> block grouped into Core Rules, Established Facts, and Recent Context.

forget_memory

id

Delete a specific memory and its incident graph edges by ID.

delete_by_context

key, value

Delete all memories matching a specific [key, value] context metadata pair.

feedback

id, useful

Record positive/negative utility feedback on a memory and adjust its salience score.

consolidate

namespace

Deduplicate memories within a namespace and promote recurring high-salience episodic memories to Semantic tier.

health

(none)

Return Memoria Cloud engine health, active memory/edge counts, and tier distribution.

stats

namespace

Return per-namespace memory count, edge count, and tier distribution.

search_code

query

Search indexed code symbol definitions and references with file paths and line numbers.

explain_code

target

Explain a symbol: definitions, workspace references, related tests, and associated architectural memories.

list_agent_wings

(none)

List specialist multi-agent wings (architect, reviewer, qa, security, ops) and active memory counts.

handoff_to_agent

from_wing, to_wing, summary, task

Record a structured context handoff between specialist agent wings.


Running Connector Tests

npm test

License

Connector package licensed under MIT. Memoria Cloud (https://memoria.kolivri.com) is a commercial hosted service by Kolivri.

Available Tools

14 tools
batch_encodeBatch Encode MemoriesAInspect

Encode multiple memories atomically in a single batch call.

Inputs: memories (required), namespace.

build_contextBuild Session ContextA
Read-onlyIdempotent
Inspect

Assemble a token-budgeted <memory_context> block combining Core rules, Semantic facts, and Episodic history for session grounding.

Inputs: query (required), mode, namespace, token_budget.

consolidateConsolidate Namespace MemoriesB
Idempotent
Inspect

Deduplicate memories within a namespace, link superseded records, and promote recurring high-salience episodic memories to Semantic tier.

Inputs: namespace (required).

delete_by_contextDelete Memories by ContextB
DestructiveIdempotent
Inspect

Delete all memories matching a specific context metadata [key, value] pair.

Inputs: key (required), value (required), namespace.

encode_memoryEncode MemoryCInspect

Store a new memory with tier classification (episodic, semantic, buffer, core), salience weight, and key-value context metadata.

Inputs: content (required), type, context, salience, +1 more.

explain_codeExplain Code SymbolA
Read-onlyIdempotent
Inspect

Explain a code symbol: where it is defined, who references it across the workspace, related tests, and associated architectural memories.

Inputs: path, depth, symbol, target, +1 more.

feedbackRecord Memory FeedbackAInspect

Record whether a recalled memory was useful and adjust its salience score.

Inputs: id (required), useful (required).

forget_memoryForget MemoryA
DestructiveIdempotent
Inspect

Delete a specific memory and its incident graph edges by ID.

Inputs: id (required).

handoff_to_agentHand Off Context to Specialist WingBInspect

Hand off task context, architectural findings, and selected memory IDs from one specialist agent wing to another.

Inputs: summary (required), task, wing, files, +5 more.

healthCheck Memoria HealthA
Read-onlyIdempotent
Inspect

Return Memoria engine health, storage status, tier distribution, and cloud sync mode.

Inputs: no named parameters.

list_agent_wingsList Specialist Agent WingsA
Read-onlyIdempotent
Inspect

List specialist multi-agent wings (architect, reviewer, qa, security, ops), their active memory counts, and recent handoffs.

Inputs: namespace.

recall_memoriesRecall MemoriesB
Read-onlyIdempotent
Inspect

Retrieve relevant memories ranked deterministically by BM25, salience, tier weight, feedback utility, and temporal freshness.

Inputs: query (required), mode, limit, namespace, +1 more.

search_codeSearch Indexed CodeA
Read-onlyIdempotent
Inspect

Search indexed code definitions and workspace source files by symbol or query, returning ranked definitions with file and line numbers.

Inputs: query (required), path, type, namespace.

statsGet Namespace StatisticsB
Read-onlyIdempotent
Inspect

Return detailed memory, edge, and tier statistics for a namespace.

Inputs: namespace (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.5/5.0

    Scored across 14 tools

    Disambiguation4/5

    Most tools map to distinct actions, but a few pairs sit close together: encode_memory vs batch_encode, forget_memory vs delete_by_context, and health vs stats (both report tier/storage metrics). Descriptions do enough to separate them, so misselection risk is moderate rather than severe.

    Naming Consistency3/5

    The dominant pattern is snake_case verb_noun (encode_memory, recall_memories, delete_by_context, list_agent_wings, handoff_to_agent), but several tools break it with bare verbs or nouns: feedback, consolidate, health, stats. Conventions are mixed, yet each name remains readable and self-explanatory.

    Tool Count5/5

    14 tools is well within a healthy range and each one covers a real capability: memory lifecycle, context assembly, maintenance, observability, code lookup, and multi-agent handoff. Nothing feels padded or redundant.

    Completeness4/5

    The memory lifecycle is largely covered (encode/batch, recall, build_context, delete, feedback, consolidate), plus code-intelligence and agent handoff extras. Minor gaps exist: no get_memory by ID, no edit/update of an existing memory's content or context metadata, and no browsing/listing of memories by namespace.

    Maintenance

    ActivityMaintained
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      A
      maintenance
      Enables AI coding agents to read and write persistent, local-first memory across sessions and projects, with semantic search, global guidelines, and optional autonomous task execution.
      -
    • F
      license
      Not graded
      quality
      A
      maintenance
      Gives AI coding agents persistent, local-first memory so they can store, semantically search, and recall decisions, preferences, error patterns, and cross-project guidelines directly during a session. It also optionally queues tasks for autonomous execution against that same memory.
      -
    • F
      license
      Not graded
      quality
      A
      maintenance
      Enables AI coding agents to store and retrieve persistent project and cross-project memories via semantic search, so decisions, preferences, error patterns, and guidelines survive across sessions. Optionally supports autonomous task execution built on the same local-first knowledge store.
      -
    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables AI agents to share a persistent, self-improving memory by ingesting sources like Obsidian vaults, code, transcripts, and session captures, then serving them back through search, classification, synthesis, and token-optimized context briefings. It lets multiple agents read and write the same knowledge base, so each session builds on the last without copy-pasting context.
      10 npm
      MIT