Mem0 MCP Server: Memories, Profiles and History

💡 No install? Use cloud.anythingmcp.com directly. Sign in, click Connectors → Mem0, paste your credentials, mint an MCP API key and you're done. No Docker, no git clone, no local server.

Mem0

A long-term memory for your AI with the Mem0 Platform: store facts from conversations per user, agent, app or session, search them by meaning, list, read and correct memories and trace how they changed, and see users, agents, profiles, processing events and usage.

Setup

Setting up takes about 2 minutes. You need a Mem0 Platform account; the free plan has monthly limits on stored memories and searches.

1. Sign in at https://app.mem0.ai and open API Keys (https://app.mem0.ai/dashboard/api-keys).
2. Create a key and copy it. It belongs to one organization and project: all memories are read and written there.
3. Paste it into MEM0_API_KEY, without the word Token. The connector sends it as Authorization: Token <key>.
4. Install the connector and run mem0_ping: it shows the organization, project and account email the key works on.

Costs: no tool is billed per call, but the plan counts usage. mem0_add_memories counts against the plan's limit of stored memories, and mem0_search_memories counts as a retrieval (search) call. mem0_get_stats shows the totals and the plan's retrieval limit.

Scoping: every memory belongs to at least one entity id: user_id, agent_id, app_id or run_id (a session). Use stable ids from your own system, such as a customer number. Search and list need a filters object with at least one of them, for example {"user_id": "alice"}, or combined: {"AND": [{"user_id": "alice"}, {"agent_id": "support-bot"}]}. Filters also take OR, NOT and the operators in, gte, lte, gt, lt, ne, contains and icontains, for example {"created_at": {"gte": "2026-01-01"}}.

Remembering: mem0_add_memories takes messages (user and assistant turns) and an entity id. Mem0 extracts the lasting facts, such as "prefers vegetarian food". Extraction runs in the background: the answer is status PENDING with an event_id, and mem0_get_event shows when it SUCCEEDED and which memories were added or changed. infer: false stores each message word for word and answers at once. metadata, expiration_date, includes, excludes and custom_instructions steer what is kept. Before storing something personal, the AI tells you what it is about to save.

Recalling: mem0_search_memories takes a natural-language query and filters and returns memory, score (relevance), categories, metadata and dates. top_k (default 10, at most 100) gives more hits, rerank: true a better order, threshold a minimum score.

Browsing and correcting: mem0_list_memories lists every memory that matches filters, newest first, with optional date range, categories and keywords; mem0_get_memory reads one; mem0_get_memory_history shows how it changed (old and new text, event ADD, UPDATE or DELETE, and the input that caused it). mem0_update_memory replaces the text, metadata or expiration_date of one memory; the AI confirms the memory and the new text with you first.

Entities, profiles and activity: mem0_list_entities lists users, agents, apps and runs with their memory counts; mem0_get_entity reads one; mem0_create_user registers a user id with metadata before any memory exists. mem0_get_user_profile returns the profile Mem0 builds for a user in the background; a 200 answer with a status and no profile means it is not ready yet. mem0_list_events and mem0_get_event show add and search operations with status (PENDING, RUNNING, SUCCEEDED, FAILED), latency and errors.

Switched off at install: mem0_delete_memory and mem0_delete_entity. Neither can be undone, and deleting an entity removes all of its memories, so a workspace admin turns them on when the team needs them.

Data formats: ids are UUIDs. Dates are ISO 8601, UTC; date filters on add and list take YYYY-MM-DD. categories are Mem0's own tags such as food, travel or professional_details.

Paging: mem0_list_memories takes page (from 1) and page_size (default 50, at most 100); mem0_list_entities pages the same way. List answers carry count, next and previous; a null next means the last page. To answer a question, the AI prefers mem0_search_memories over paging through everything.

Limits: plans limit stored memories and search calls per month; mem0_get_stats shows total_retrieval_events_limit.

Common errors: 400 means missing or invalid filters (search and list reject a top-level user_id; it belongs in filters), no entity id on add, or a bad date format; 401 a wrong or revoked key; 404 an unknown memory, entity or event; 429 the rate or plan limit is reached (the AI waits instead of retrying in a loop).

The connector does not delete all memories in bulk and does not run Mem0's open-source library; it works with the hosted Mem0 Platform only.

Prerequisites: MEM0_API_KEY

Install the connector locally

curl -fsSL https://raw.githubusercontent.com/HelpCode-ai/anythingmcp/main/docker-compose.quickstart.yml -o docker-compose.yml
printf 'JWT_SECRET=%s\nENCRYPTION_KEY=%s\n' "$(openssl rand -hex 32)" "$(openssl rand -hex 32)" > .env
docker compose up -d

Open http://localhost:3000/connectors/store, pick Mem0 and paste in the env vars listed under Prerequisites.

Available tools

Example prompts

• "Remember for user alice that she is vegetarian and allergic to nuts."
• "What do we know about the food preferences of user alice? Search her memories."
• "Did the memory extraction from my last add succeed, and which memories did it create?"
• "The memory that says alice lives in Lyon is outdated: she moved to Turin. Correct it and show me its history."
• "Which users have the most memories, and how close are we to the plan's search limit?"

Next steps

• Connect to Claude
• Connect to ChatGPT