Mem0
by Mem0
Give your AI agent persistent memory across sessions. 11 tools for saving, searching, updating and deleting memories, OAuth or API-key sign-in, three of the eleven are deletes.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect Mem0 via MCP
https://mcp.mem0.ai/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Mem0 Tools & Capabilities (11)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- Deletes are irreversible. No soft delete, no trash, no restore path is published for memories removed through the API.
- The delete confirmation is client-side. Mem0's tool descriptions instruct the model to confirm a scope; the server does not enforce it.
user_id=*is documented. Each delete filter accepts a wildcard meaning all users, agents or apps in scope.- No safety annotations were readable. The gated endpoint returned no
tools/list, so noreadOnlyHintordestructiveHintcould be checked. - No
permissionslabel in Anthropic's directory. The read-or-write posture is simply not asserted. - Monthly request allowances. Mem0's pricing page lists 10,000 add and 1,000 retrieval requests a month on the free Hobby tier, 50,000 and 5,000 on Starter at $19 a month, and 500,000 and 50,000 on Pro at $249 a month, all retrieved 2026-08-19. An agent saving memories on its own initiative spends the add allowance without a human decision per call.
- Free-tier content trains Mem0's models. Mem0's privacy policy states inputs and user-generated content from Free Plan users are used for model training and Paid Plan users' are not.
- Batch delete caps at 1,000. Mem0's batch endpoint accepts at most 1,000 memory IDs per call; the filter-based delete has no equivalent ceiling.
- Not every message becomes a memory. Mem0's classifier discards definitional and general- knowledge content, so an agent told to remember something may store nothing.
- The endpoint needs its trailing slash.
https://mcp.mem0.ai/mcpanswered our probe with a 307 redirect to/mcp/before returning the 401; a client that does not follow redirects on POST may report a confusing failure.
Frequently asked questions
What exactly does `delete_all_memories` delete?
Everything inside one scope, and the scope is an argument rather than a default. Mem0's docs describe the tool as a bulk delete of all memories in scope, where scope means a `user_id`, `agent_id`, `app_id` or `run_id` filter. The underlying API rejects a call with no filter at all, but each filter separately accepts `*` as a documented wildcard meaning every user, agent or app.
Can I undo a Mem0 delete made by an agent?
No. Mem0 publishes no soft delete, no trash and no restore path for memories removed through the API, and its delete guide's own verification step is to search again and confirm the memory is gone. If you need a memory to stop surfacing without being destroyed, Mem0 documents `expiration_date` instead, which hides a record from list-shaped reads while leaving it in storage.
Does Mem0 ask me to confirm before an agent deletes memories?
Not at the server. The confirmation Mem0 documents lives in the tool descriptions, which tell the model to act on a confirmed scope, so the enforcement is whatever your MCP client does with a write tool. Claude and most clients prompt before a tool call by default, but that is a client behaviour and not a Mem0 platform block. Nothing on the server side refuses an unconfirmed delete.
What is an entity in Mem0, and what does `delete_entities` remove?
An entity is one of four scoping identifiers: a user, an agent, an app or a run. Mem0's API reference describes deleting an entity as removing that entity and all its memories, so it is broader than deleting memories alone. These are the partitioning identifiers, not the people and places that Graph Memory extracts, which Mem0's docs call out as a separate concept.
Does Mem0 train its AI models on my memories?
It depends on your plan, and Mem0 states the split plainly. Its privacy policy says inputs, prompts and user-generated content from Free Plan users are used to train its AI models, and that it does not train on the same data from Paid Plan users. Since a memory store holds whatever an assistant chose to remember about someone, that distinction is worth checking before connecting on the free tier.
Does the Mem0 MCP server work with an API key instead of browser sign-in?
Yes. Mem0 documents two ways in: the browser OAuth flow most clients trigger on first tool use, and sending a Platform API key as a bearer token for clients without browser sign-in or for headless environments like CI. Both reach the same hosted server. Our own probe saw only the OAuth challenge, since we never authenticated.
How many Mem0 requests do I get before I hit a limit?
Mem0's pricing page publishes a monthly allowance per plan, split between writes and reads. The free Hobby tier lists 10,000 add requests and 1,000 retrieval requests a month, Starter lists 50,000 and 5,000, and Pro lists 500,000 and 50,000. An agent that saves memories automatically consumes the add allowance without anyone deciding to spend it.
Sources
- Mem0 MCP documentation — https://docs.mem0.ai/platform/mem0-mcp (fetched 2026-08-19, HTTP 200, no redirect) · retrieved 2026-08-19
- Mem0 delete memories API reference — https://docs.mem0.ai/api-reference/memory/delete-memories (fetched 2026-08-19) · retrieved 2026-08-19
- Mem0 delete user (entities) API reference — https://docs.mem0.ai/api-reference/entities/delete-user (fetched 2026-08-19) · retrieved 2026-08-19
- Mem0 memory operations, delete — https://docs.mem0.ai/core-concepts/memory-operations/delete (fetched 2026-08-19) · retrieved 2026-08-19
- Mem0 memory expiration — https://docs.mem0.ai/platform/features/memory-expiration (fetched 2026-08-19) · retrieved 2026-08-19
- Mem0 entity-scoped memory — https://docs.mem0.ai/platform/features/entity-scoped-memory (fetched 2026-08-19) · retrieved 2026-08-19
- Mem0 events API reference — https://docs.mem0.ai/api-reference/events/get-events and https://docs.mem0.ai/api-reference/events/get-event (fetched 2026-08-19) · retrieved 2026-08-19
- Mem0 platform FAQs — https://docs.mem0.ai/platform/faqs (fetched 2026-08-19) · retrieved 2026-08-19
- Mem0 documentation index — https://docs.mem0.ai/llms.txt (fetched 2026-08-19) · retrieved 2026-08-19
- Mem0 pricing — https://mem0.ai/pricing (fetched 2026-08-19) · retrieved 2026-08-19
- Mem0 privacy policy — https://mem0.ai/privacy, which redirects to https://mem0.ai/privacy-policy (fetched 2026-08-19) · retrieved 2026-08-19
- Archived
mem0ai/mem0-mcpREADME — https://raw.githubusercontent.com/mem0ai/mem0-mcp/main/README.md (fetched 2026-08-19) · retrieved 2026-08-19 - Live auth probe — HTTP 401 with a
www-authenticateBearer challenge andresource_metadatapointer, 2026-08-19 · retrieved 2026-08-19 - RFC 9728 protected-resource descriptor — https://mcp.mem0.ai/.well-known/oauth-protected-resource (fetched 2026-08-19;
/mcppath-insert and path-append forms both 404) · retrieved 2026-08-19 - RFC 8414 authorization-server metadata — https://mcp.mem0.ai/.well-known/oauth-authorization-server (fetched 2026-08-19) · retrieved 2026-08-19
- Anthropic connector directory — https://claude.ai/directory/3119bc71-0dfd-48f8-a764-c2fca895325c (snapshot 2026-08-16) · retrieved 2026-08-16
docs.mem0.ai/robots.txtcarriesContent-Signal: ai-train=yes, search=yes, ai-input=yes(fetched 2026-08-19) — Mem0 expressly permits AI synthesis of these pages. · retrieved 2026-08-19
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Developer Tools
- Developer
- Mem0
- Tools
- 11
- Domain
- mcp.mem0.ai
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.