Memory Service MCP Server by Doobidoo: Setup for Claude, Cursor and More

• Saves a memory with tags

  Saves a piece of information as a memory. You can add tags and metadata so you can find it again.

  Try: “Remember that we chose Postgres for the billing service.”

  Write · memory_store

• Saves a whole conversation as one memory

  Stores a full multi-turn conversation as a single unit, so later searches find the context together.

  Try: “Save this conversation as a session memory.”

  Write · memory_store_session

• Finds memories

  Searches stored memories by meaning, exact words or tags, with filters. It is the main way to recall something.

  Try: “What did we decide about the billing database?”

  Read · memory_search

• Browses memories

  Lists memories page by page, with optional filters.

  Try: “List my 20 most recent memories.”

  Read · memory_list

• Deletes memories by filter

  Deletes memories that match filters you combine, such as tags, dates or content hash.

  Try: “Delete every memory tagged scratch.”

  Delete · memory_delete

• Removes duplicates

  Finds duplicate memories and removes them.

  Try: “Clean up duplicate memories.”

  Delete · memory_cleanup

• Checks the database

  Reports whether the memory database is healthy and gives its statistics.

  Try: “Is my memory store healthy?”

  Read · memory_health

• Shows cache and speed stats

  Returns caching and start-up timing numbers for the server, for monitoring.

  Try: “Show memory server performance stats.”

  Read · memory_stats

• Edits a memory's tags and metadata

  Changes the tags or metadata of a memory without recreating it. The content and its embedding stay as they were.

  Try: “Add the tag billing to that memory.”

  Write · memory_update

• Runs memory tidy-up

  Runs, checks or schedules consolidation, which merges and prunes old memories to keep the store lean.

  Try: “Run a consolidation pass on my memories.”

  Delete · memory_consolidate

• Imports documents or folders

  Reads a PDF, text, Markdown or JSON file, or a whole folder, from your computer and stores it as memories.

  Try: “Ingest the docs in ~/notes into memory.”

  Write · memory_ingest

• Extracts learnings from Claude Code sessions

  Reads Claude Code session transcripts from your computer and stores the decisions, bugs and conventions it finds.

  Try: “Harvest learnings from this session.”

  Write · memory_harvest

• Rates and analyzes memory quality

  Rates a memory, shows its quality numbers, analyzes quality across the store, or runs maintenance.

  Try: “Rate that memory as good.”

  Write · memory_quality

• Explores links between memories

  Follows connections between memories: finds connected ones, shortest paths, subgraphs and entity profiles.

  Try: “Show memories connected to the billing decision.”

  Write · memory_graph

• Lists contradicting memories

  Lists memories that contradict each other and have not been resolved.

  Try: “Do any of my memories conflict?”

  Read · memory_conflicts

• Settles a memory conflict

  Resolves a conflict by choosing which memory wins.

  Try: “Keep the newer memory in that conflict.”

  Delete · memory_resolve

• Records a mistake and its fix

  Stores what went wrong and the correct action. A repeated mistake raises its failure count.

  Try: “Note that deploying without migrations broke staging.”

  Write · mistake_note_add

• Finds past mistakes

  Searches mistake notes by meaning, to check for known pitfalls before a task.

  Try: “Any known pitfalls for database migrations?”

  Read · mistake_note_search

• Edits a mistake note

  Changes the fields of an existing mistake note. Changing the content replaces the note.

  Try: “Update the fix on that mistake note.”

  Delete · mistake_note_update

• Deletes a mistake note

  Deletes a mistake note by its hash. It only deletes memories of the mistake type.

  Try: “Delete that mistake note.”

  Delete · mistake_note_delete

• Lists flagged memories

  Lists memories held back because they contradict active beliefs, so you can review them.

  Try: “Show quarantined memories.”

  Read · get_quarantined_memories

• Releases a flagged memory

  Takes a memory out of quarantine when it turns out to be valid.

  Try: “Release the quarantined memory about staging.”

  Write · unquarantine_memory

• Saves end-of-session learnings

  Stores the decisions, errors and corrections from a short-lived agent session as structured notes.

  Try: “Commit what we learned this session.”

  Write · commit_session_legacy

• Builds a start-up profile for an agent

  Builds a profile of past behavior for a fresh agent. It can reset session counters, so it counts as a write.

  Try: “Get my bootstrap profile.”

  Write · get_bootstrap_profile

• Returns a setup guide for your app

  Returns the integration guide for a given type of client.

  Try: “How do I set up memory for Cursor?”

  Read · get_onboarding_guide

• Distills insights from memories

  Uses a language model to pull insights out of existing memories in a batch.

  Try: “Distill insights from my project memories.”

  Write · memory_distill

• Maps entities in your memories

  Returns a knowledge map of entities with matching snippets for each. It needs an entity graph that already has data.

  Try: “Explore what I know about the billing service.”

  Read · memory_explore

• Shows one entity in depth

  Returns the full ranked detail for a single entity from the knowledge map.

  Try: “Show everything about the billing service entity.”

  Read · memory_detail

Watch out for

• It can delete memories

  memory_delete, memory_cleanup and mistake_note_delete remove stored memories, and memory_consolidate can merge or prune them. Back up the database before long sessions. github.com

• It reads files from your computer

  memory_ingest reads documents from any folder you name, and memory_harvest reads Claude Code transcripts. Check the paths before approving. github.com

• The HTTP mode can run without a login

  The README starts the web server with MCP_ALLOW_ANONYMOUS_ACCESS=true. Keep it on 127.0.0.1 and never expose it to a network that way. github.com

• It downloads a model on first run

  The embedding model comes from an Amazon S3 address the first time it runs. Allow this on a restricted network or install the model in advance. github.com

Security advisories