PlugWiki
hippo-memory logo

Hippo Memory MCP Server: Setup for Claude Code, Cursor and Other Apps

OfficialLocal
Getnpmhippo-memorykitfunso/hippo-memory

Lets your coding assistant save and recall project memories that stay on your machine and carry across apps.

Hippo gives Claude, Cursor, Codex and other coding assistants one shared long-term memory for your project. Ask the assistant to remember a fact, and it can bring it back in a later session, even in a different app. The memories sit in a SQLite file on your machine.

The MCP server runs locally with no account or API key. It reads and writes the project's memory store, and three of its tools can delete or change memories, copy them to a shared store, or scan your git history. Memories fade over time unless they are recalled or pinned.

Good for

  • Developers who switch between Claude Code, Cursor and Codex and want one memory
  • People who want lessons from past fixes and reverts saved automatically
  • Teams who need project memory stored locally, not in a hosted service

Works in

12 should work

Tap an app to see how to set it up and any known problems.

Install hippo-memory MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the hippo-memory MCP server to Claude Code. Run this command:
    
    claude mcp add hippo-memory -- npx -y hippo-memory mcp
    
    Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

The command must end in mcp: plain hippo-memory is the command-line tool, not the MCP server. Run hippo init inside a project afterwards to create its .hippo/ store. Without it, the first tool call creates a global store in ~/.hippo. Run hippo doctor to check the install.

Built from the package data we hold. Not yet tested by us.

Tools

  • Searches your saved project memories for a topic

    Finds the memories that match a question and returns them ranked by relevance, strength and recency, within a size limit you can set. It can also add the active task notes and the last session handoff, which helps at the start of a session.

    Try: “Recall what we learned about the payment webhook retries.”

    Read · hippo_recall

  • Builds an ordered context window for one session

    Takes a session ID and returns that session's recent messages untouched, plus short summaries standing in for older ones, trimmed to a token budget. It drops the weakest items first, not the oldest.

    Try: “Assemble the context for session abc123 within 4,000 tokens.”

    Read · hippo_assemble

  • Opens a summary to see the memories behind it

    When a recalled item is a summary, this returns the original memories it replaced, one level down or more. Use it to recover the detail a summary left out.

    Try: “Show me the original notes behind that summary.”

    Read · hippo_drill

  • Saves a new memory

    Stores a short fact for later sessions. You can mark it as an error, which keeps it twice as long, pin it so it never fades, or tag it. Unpinned memories fade unless they are recalled.

    Try: “Remember that the staging database needs the VPN on.”

    Write · hippo_remember

  • Says whether the last recalled memories helped

    Records if the memories you just got were useful. Good ratings slow a memory's fade and bad ones speed it up.

    Try: “Mark those memories as helpful.”

    Write · hippo_outcome

  • Loads memories for what you are working on now

    Reads the current git state to guess the task, then returns matching memories with the active task notes, the session handoff and the recent session trail. Memories from private sources are left out unless you ask for that scope.

    Try: “Load the context for this branch.”

    Runs code · hippo_context

  • Shows the health of the memory store

    Returns counts, memory strengths, memories at risk of fading and the time of the last consolidation.

    Try: “How healthy is the memory store?”

    Read · hippo_status

  • Turns recent git fixes and reverts into memories

    Scans recent commits for fix, revert, bug, refactor and perf patterns and saves lessons from them. You choose how many days back to scan, and it skips duplicates.

    Try: “Learn from the last 14 days of commits.”

    Runs code · hippo_learn

  • Lists memories that contradict each other

    Returns the open conflicts, which are pairs of memories that disagree and need a decision.

    Try: “Are there any conflicting memories?”

    Read · hippo_conflicts

  • Settles a conflict by keeping one memory

    Keeps one memory from a conflict and weakens the other. With the forget option it deletes the loser, and with the reject option it also blocks that value from coming back.

    Try: “Resolve conflict 7 and keep memory m_42.”

    Delete · hippo_resolve

  • Copies a memory to the global store

    Shares one memory so other projects on this machine can use it. It scores how well the memory transfers and refuses low scores unless you force it.

    Try: “Share memory m_42 with my other projects.”

    Write · hippo_share

  • Lists projects that added to the global store

    Returns every project that has contributed memories to the shared global store.

    Try: “Which projects share memories here?”

    Read · hippo_peers

Read from the source code

Permissions

high risk · 5 things to watch

High risk

The assistant can save, rate and delete memories, copy them into a store shared by every project on your machine, and scan your git history. Our advice: back up the .hippo/ folder before letting an assistant run conflict resolution, because the forget option deletes memories.

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • hippo_resolve can delete memories

    With the forget option it deletes the losing memory, and with the reject option it also blocks that value from returning. Ask the assistant to confirm before it resolves conflicts. github.com

  • hippo_learn reads your git history

    It scans recent commits and saves lessons from them as memories. Commit messages end up in the memory store and can be recalled in other projects if shared. github.com

  • Shared store crosses projects

    hippo_share copies a memory to a global store that other projects can read. Do not share memories that hold client or secret details. github.com

  • Memories fade unless recalled

    Unpinned memories decay over time, so a fact the assistant saved can drop out of results later. Pin the ones that must stay. github.com

  • hippo init adds hooks and a daily job

    Running hippo init writes hooks into your agents' settings and sets up a daily 6:15am run. With ANTHROPIC_API_KEY set, the sleep step that runs at session end sends memory text to Anthropic; the README says an extraction setting in .hippo/config.json turns that off. github.com

Security advisories

No published security advisories as of Nov 1, 2026.

Technical details
Turning on read-onlyNo read-only switch is documented for the MCP server; every tool is registered by default.
Talks toRecall makes no network call by default, per the README. The sleep step sends memory text to Anthropic when ANTHROPIC_API_KEY is set
Sign-in and keysNo account or API key. Memories are stored in a SQLite file in the project's .hippo folder, or in ~/.hippo for the global store.

Questions

Is Hippo Memory official?

Yes, in the sense that the MCP server ships inside the Hippo project's own repo, built by the same author (kitfunso). Hippo is an independent project, not made by Anthropic, OpenAI or Cursor.

Is Hippo Memory free?

Yes. The MCP server and the rest of the core are MIT-licensed and the license file ships in the npm package. The README describes a separate planned commercial edition for larger companies.

Does it need an API key?

No. The README says search runs on plain keyword ranking with no model and no network call by default. Only optional features, such as embeddings, rerankers or the session-end sleep step with ANTHROPIC_API_KEY set, call out to other services.

Where are my memories stored?

In a .hippo/ folder inside each project, as a SQLite file with readable markdown copies. The first tool call with no store anywhere creates a global store in ~/.hippo.

Can it run read-only?

No switch for that is documented. All 13 tools are registered, including the ones that save, share and delete memories.

Which apps does it work with?

The README names Cursor, Windsurf, Cline and Claude Desktop for the MCP server, and the project's own setup also covers Claude Code, Codex, OpenCode, OpenClaw and Pi. Pick your app in the install box above.

What is the difference between hippo-memory and hippo-memory mcp?

The mcp ending starts the MCP server. Without it the same package is the command-line tool, so an MCP client config that leaves it off will not connect.

Can it delete my memories?

Yes. The conflict resolution tool can delete the losing memory when told to forget it. Everything else only adds, rates or reads memories.

Information

updated Nov 2, 2026

Publisher
Official
Package
hippo-memory
Registry
npm
Transport
stdio
Auth
None
Version
2.2.0
License
MIT
Last commit
Oct 29, 2026
Open issues
4

WebsiteGitHubnpmChangelogIssues

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 3
  1. Hippo README · github.com
  2. tools.ts at v3.2.0 · github.com
  3. memory-tools.ts at v3.2.0 · github.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 2, 2026.