Ratary Memory Mcp
About
Ratary is an open-source AI Brain Platform. The Memory MCP server (ratary) gives coding assistants durable, owner-scoped memory across sessions — hybrid search, knowledge graph traversal, and token-efficient context assembly.
Details
- Author
- ontorata
- Downloads
- 275
- Categories
- AI, Knowledge Base, Other, Search, Developer Tools
Jump to
- Durable, owner-scoped memory across sessions
- Hybrid search and knowledge graph traversal
- Token-efficient context assembly
- Self-hostable with multiple database backends
- Multiple transport options: stdio, REST proxy, Streamable HTTP
Setting up with Highlight
This MCP is not yet compatible with Highlight’s one-click setup. However, you can still use it with Highlight by following these steps:
- Download and install Highlight from highlightai.com/download
- Navigate to the plugins tab and select "Add Custom Plugin"
-
Configure the plugin with the settings below
Plugin Name
Ratary Memory McpCommand (node, npx, python, etc.)Please refer to the README for specific instructions on how to obtain API keys or other required environment variables.
- Enable "Start Automatically" if you want the plugin to start when Highlight launches
From the repository
Install via npm (@ratary/mcp-server) for a REST proxy with 6 tools, or clone the repository for local stdio with 28 tools. Self-host using D1, Postgres, Supabase, MariaDB, or Docker. An optional remote Streamable HTTP transport is available when deployed.
get_capabilities
Return the Ratary deployment capability manifest (protocol version, tool count, limits, feature flags). Read-only and idempotent; requires RATARY_API_KEY. Use at session start to discover what this deployment supports. Do not use for recalling memories — use search_memory or get_context instead. Returns a JSON capability object (no side effects).
search_memory
Search persistent coding memories with hybrid ranking and return ranked hits (id, title, summary, relevance). Read-only; does not create or modify memories. Requires RATARY_API_KEY. Use when you need candidate memories matching a keyword/question before answering. Prefer get_memory when you already have a memory UUID. Prefer get_context when you need token-bounded context assembled for a task (not a raw hit list). Returns JSON search results; empty list means no matches.
save_memory
Create a new persistent coding memory (title + markdown body) scoped to an optional project and tags. Side effect: writes a new record via the Ratary REST API; not idempotent (each call creates another memory). Requires RATARY_API_KEY. Does not overwrite existing memories — use get_memory/search_memory first if updating. Use after decisions, handoffs, or durable facts the agent should recall later. Do not use for ephemeral chat notes that should not persist. Returns the created memory JSON including id/codename/timestamps.
get_memory
Fetch one memory by UUID, including full content and metadata. Read-only and idempotent; requires RATARY_API_KEY. Use when you already know the memory id (from search_memory or a prior save_memory). Do not use for keyword discovery — use search_memory. Do not use to assemble multi-memory task context — use get_context. Returns the memory JSON or an API error if the id is missing.
get_context
Assemble token-efficient ranked context for a coding task from persistent memories (summaries by default). Read-only; does not write memories. Requires RATARY_API_KEY. Use at the start of implementation or when answering with organizational memory. Prefer search_memory when you need raw ranked hits to inspect individually. Prefer get_memory for a known UUID. Returns markdown/JSON context suitable for injecting into the agent prompt.
list_agent_clients
List certified external agent client profiles registered with this Ratary deployment (name, transport, status). Read-only and idempotent; requires RATARY_API_KEY. Use for ecosystem/discovery questions about which clients are supported. Do not use for memory CRUD or retrieval — use save_memory / search_memory / get_context. Returns a JSON list of client profiles (no side effects).
Claude Desktop / Cursor
Paste into your MCP client config file to install this server.
{
"mcpServers": {
"ratary memory mcp": {
"ratary": {
"command": "npx",
"args": [
"-y",
"@ratary/mcp-server"
],
"env": {
"RATARY_BASE_URL": "https://ratary.ontorata.com",
"RATARY_API_KEY": "aic_..."
}
}
}
}
}McpServers
{
"ratary": {
"command": "npx",
"args": [
"-y",
"@ratary/mcp-server"
],
"env": {
"RATARY_BASE_URL": "https://ratary.ontorata.com",
"RATARY_API_KEY": "aic_..."
}
}
}Category:Memory ·Transport:stdio (local) + Streamable HTTP (remote, opt-in)
Repository:github.com/ontorata/ratary
Listing:submit tomcpservers.orgwith linkhttps://github.com/ontorata/ratary/tree/main/MCP
Persistentcoding memoryfor AI assistants — save, search, build token-efficient context, knowledge graph traversal, and multi-client sync. Works withCursor,Claude Code,Roo,Cline,Gemini CLI, and remote MCP hosts (ChatGPT App URL when deployed).
Ecosystem:Built byOntorata. This doc coversRatary Memory MCP(idratary).Ontorata MCPandOntorata Studioare separate repos.
npm:Hosted REST proxy —@ratary/mcp-server(@rataryorg). Full stdio (30 tools) requires cloning this repo.
Tool registry SSOT:src/capabilities/mcp-tool-names.ts
Configureone SQL metadata providerfirst — seeCONFIGURATION — SQL metadata store.
git clone https://github.com/ontorata/ratary.git cd ratary npm install cp .env.example .env # Set SQL_PROVIDER + matching credentials (D1, DATABASE_URL, or MARIADB_CONNECTION_STRING) npm run db:migrate # D1 only — use db:apply-postgres-schema for Postgres / Supabase
REST dev server:npm run dev→http://localhost:9876(Swagger/docs). Override withPORTin.env.
Writes.cursor/mcp.jsonand.mcp.jsonautomatically.
Seedocs/examples/mcp/cursor.mcp.json.example— replaceREPO_PATHwith your clone path.
{ "mcpServers": { "ratary": { "command": "npx", "args": ["-y", "tsx", "REPO_PATH/src/mcp/stdio.ts"], "cwd": "REPO_PATH" } } }
Reload MCP in your IDE. No API key needed in Cursor when using direct D1 mode.
For teams using a deployed Ratary REST endpoint:
npm install -g @ratary/mcp-server export RATARY_BASE_URL=https://ratary.ontorata.com export RATARY_API_KEY=aic_... ratary-mcp
Config example:docs/examples/mcp/remote-api.mcp.json.example
Endpoint:https://your-host/mcp(Beareraic_...or OAuth when enabled).
CI smoke:tests/transport/remote-mcp-chatgpt-smoke.test.ts(ChatGPT-style initialize payload).
Details:GUIDE — ChatGPT·CONFIGURATION — Tier 4
Tool failures never surface as MCP protocol errors. Any handler exception — and any invalid/missing argument — returns a structured tool result (isError: true) whose text is parseable JSON:
{ "error": "<message>", "retryable": false }
retryableis aclient-behavior hint, not a statement about the current implementation:
- retryable: true— idempotent reads (search_memory,get_memory,get_context,build_prompt,list_,traverse_relations,get_capabilities,negotiate_capabilities,get_compression_status,sync_pull,sync_status) failing transiently. Retry with a short bounded backoff (2–3 attempts).
- retryable: false— all mutations (save_memory,update_memory,delete_memory,link_memories,toggle_favorite,archive_memory,register_agent,submit_signal,sync_push) plusrun_stewardship(one run may partially succeed across sub-stages; automatic retry risks double maintenance). Also every deterministic failure (validation, not-found, auth) on any tool — retrying identical input cannot succeed.
- Never blind-retry a write on an ambiguous timeout— a silent success followed by a retry creates duplicates. Either pass arequest_id(below) so the retry is safe, or continue the turn and reconcile on the nextsearch_memory/recall.
- Treat memory as best-effort context, not a hard dependency.If a call fails, proceed with the context you already have and try again next turn. A missed write is recoverable; a crashed agent turn is not.
- Classification source of truth:src/transport/mcp/mcp-tool-retry-classification.ts· contract regression suite:tests/mcp-error-contract/.
save_memoryaccepts an optionalrequest_id(UUID, same style assubmit_signal'ssignal_id). Generate one per logical create and reuse it on every retry of that create:
- First call with a givenrequest_idcreates the memory normally.
- Any retry with the samerequest_id— including after an ambiguous timeout — returns theoriginal memoryas a success, enriched with"duplicate": true, "replayed": true. No second row is ever created, even if the first attempt crashed mid-write.
- sync_pushcreate items get the same protection automatically, keyed by the item'smemory_id— re-pushing a batch replays instead of duplicating.
Idempotency is guaranteed while the intent record exists.Completed intent records are pruned afterWRITE_INTENT_TTL_DAYS(default 30 days) as a cleanup policy — a retry arriving after that window may create a duplicate. Cleanup never deletes an unresolved (claimed-without-result) intent; those are kept and surfaced in stewardship findings. Without arequest_id, behavior is unchanged: identical saves create distinct memories.
Design: ADR-067 · contract suite:tests/idempotent-writes/.
SubmitRatary Memory MCPto public directories using the copy-paste pack inMCP/submission/.
Repo-local metadata for tooling:server.json(stdio + npm + remote flags). Registry publish usessubmission/official-registry.server.json.
Boundary:ListRatary Memory MCP(ratary) only — notOntorata MCPorOntorata Studio.
Persistent shared memory for AI agents. Hybrid search (pgvector + tsvector), knowledge graph, cognitive scoring - 97.2% Recall@10 on LongMemEval
Art provenance intelligence: 282,731-node knowledge graph over Getty Provenance Index, Knoedler stock books, and wartime-restitution records — every answer cited, custody gaps named. Free keyless remote: https://provenio.art/api/mcp
Neo4j graph database server (schema + read/write-cypher) and separate graph database backed memory
Remote MCP for commodities research — knowledge graph, COT positioning, natural gas storage/weather, futures quotes, and private stories/journals. Requires Arc Research Starter + bearer token.
A task management system for LLM agents to manage projects, tasks, and knowledge using a Neo4j database for complex workflow automation.
MCP server for live NFL and fantasy football data, powered by gamedai Scout: player grades, start/sit calls, and a grounded football knowledge graph.
Mines git history for file co-change patterns into a queryable knowledge graph via MCP.
Query a hybrid graph (Neo4j) and vector (Qdrant) database for powerful semantic and graph-based document retrieval.
A knowledge graph-driven persistent memory layer for coding agents and LLM workflows.
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.
