M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
H

Hivemind

Community
One brain for all your agents
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #280 of 438
Stars
★ 1.6k Very popular
Transport
stdio (local process)
Runtime
Node.js >=22.0.0
Credentials
API key / credential required
License
Apache-2.0
Last commit
Tools
3
48FMRS · D

Hivemind provides powerful collaboration for team coding agents through shared memory and skill reuse. However, it depends on cloud storage and exposes the risk of data sharing visibility.

Strongest · Documentation 12/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
8/20
Maintenance
10/20
Documentation
12/20
Setup experience
10/20
Why each score
Reliability 8/20
Evidence: The repo contains a substantial test suite (vitest) and CI workflows, but they are not executed here, and the README does not detail test coverage. Source code implements an MCP server (server.js) with tools like hivemind_search, but actual behavior is unverified. Deductions: no execution, limited validation of tool behavior consistency, and limited error handling details.
Security and permissions 8/20
Evidence: Source claims SQL escaping, credentials stored with 0600 perms, device-flow login without env tokens. Data collection notice and privacy controls (e.g., HIVEMIND_CAPTURE=false) are present. Deductions: data is sent to external Deeplake cloud service, insufficient evidence of user confirmation for data capture; MCP server may auto-register with multiple clients, least-privilege scoping not fully assessed. No red-line issues found.
Maintenance 10/20
Evidence: Active repository with high stars, continuous commits and releases, Apache-2.0 license. Deductions: issue response times and dependency update frequency not verified, but overall maintenance signals are strong.
Documentation 12/20
Evidence: README is very detailed, covering installation, configuration, usage examples, security/storage, and benchmarks. Deductions: missing per-tool parameter documentation for MCP tools (e.g., hivemind_search parameters), no explicit MCP client config example, limited troubleshooting details.
Setup experience 10/20
Evidence: One-line install command and per-assistant installation instructions are provided. Deductions: Requires external account and token, restart after install, manual configuration potential; no direct client config snippet provided, and full setup flow is not validated.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesUses the networkConnects to a database

Best for

  • Team-based coding agent environments.
  • Scenarios requiring long-term memory and cross-session knowledge sharing.
  • Users who want to automatically extract skills from historical traces.

Not for

  • Single-user scenarios without need for shared memory.
  • Environments requiring complete data isolation without external storage.

Required permissions

  • Captures user prompts, tool calls, tool responses, and assistant responses.
  • Stores data in Deeplake cloud or BYOC storage.

Risks and side effects

  • Data is shared across all users in a workspace, posing potential privacy leaks.
  • Relies on cloud services, so data leaves local machines unless using BYOC.

Setup

Before you start

Runtime：Node.js >=22.0.0

HIVEMIND_TOKEN requiredsecret API token for Deeplake sign-in, from your account settings at https://deeplake.ai; required for headless/CI installs.
Other optional settings (19)
HIVEMIND_ORG_ID optional Organization ID, normally auto-set by hivemind login.
HIVEMIND_WORKSPACE_ID optional Workspace name, defaults to default.
HIVEMIND_API_URL optional Deeplake API endpoint, defaults to https://api.deeplake.ai.
HIVEMIND_TABLE optional SQL table for summaries and the virtual filesystem, defaults to memory.
HIVEMIND_SESSIONS_TABLE optional SQL table for per-event session capture, defaults to sessions.
HIVEMIND_MEMORY_PATH optional Path that triggers file-operation interception, defaults to ~/.deeplake/memory.
HIVEMIND_CAPTURE optional Set to false to disable capture entirely (read-only mode).
HIVEMIND_CAPTURE_ONLY_CLI optional Set to true to capture only interactive CLI sessions, skipping Claude Agent SDK sessions.
HIVEMIND_SKILLIFY_EVERY_N_TURNS optional Assistant turns between auto skill-mining attempts, defaults to 20.
HIVEMIND_SUMMARY_EVERY_N_MSGS optional Captured events between periodic session summaries, defaults to 50.
HIVEMIND_SUMMARY_EVERY_HOURS optional Time-based summary cadence in hours, defaults to 2.
HIVEMIND_WIKI_WORKER optional Set to 1 to disable the background session-summary worker.
HIVEMIND_GRAPH_ON_STOP optional Set to 0 to disable the code-graph rebuild on Stop/SessionEnd.
HIVEMIND_EMBEDDINGS optional Set to false to force lexical-only (BM25) retrieval.
HIVEMIND_PROACTIVE_RECALL_DISABLED optional Set to 1 (or HIVEMIND_PROACTIVE_RECALL=0) to disable proactive recall.
HIVEMIND_RECALL_MIN_OVERLAP optional Minimum keyword overlap for lexical proactive recall, defaults to 2.
HIVEMIND_RECALL_TIMEOUT_MS optional Hard cap on the proactive-recall synchronous search path in ms, defaults to 1000.
HIVEMIND_DEBUG optional Set to 1 for verbose hook debug logs.
HIVEMIND_RULES_TABLE optional Table name for team rules, defaults to hivemind_rules.

Install with: npm i -g @deeplake/hivemind && hivemind install. The installer detects all supported assistants on your machine and wires up the hooks. For Claude Cowork, run hivemind claude_cowork install.

Check that it works

Run `hivemind status` after install to confirm each assistant is wired up; check that hivemind_search, hivemind_read and hivemind_index appear in the client's tool list, then ask a natural-language search question — a hit from shared team memory proves the connection.

Troubleshooting

  1. If agents feel sluggish, check the model; Hivemind makes many small tool calls, so a fast model is recommended.
  2. If the integration doesn't take effect after install, restart your assistants.
  3. If no data is captured, verify `HIVEMIND_CAPTURE` env var is not set to false accidentally.

Things to try

Once connected, you can ask your AI assistant things like:

  • Search traces for authentication bugs we've solved
  • What did we decide about the API design?
  • Read the session summary about that migration
  • Index this project's codebase so teammates can recall it

Tools 3

hivemind_search read-only
Search across team traces and skills.
hivemind_read read-only
Read a specific trace or skill from shared memory.
hivemind_index writes
List what is available for retrieval from shared memory.

Use cases

Share project knowledge across all agents in a team.
Automatically convert recurring solutions into reusable skills.
Reduce API costs by avoiding re-deriving prior work.

Supported clients

Claude Desktop

Listed from the project's documentation, not tested by this site.

Overview

Hivemind is an auto-learning, cloud-backed shared brain for agents like Claude Code, OpenClaw, Codex, Cursor, Hermes, and pi. It captures every session's prompts, tool calls, and responses as structured traces in Deeplake, mines patterns, and codifies them into reusable skills that propagate across all agents in your team. It supports hybrid lexical + semantic search and provides an SQL-backed virtual filesystem.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 62.9k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision e054fb161de7 Data synced 2026-10-12 Read the FMRS scoring method