Skip to content
MCP server

driftnet

Local catalogue that catches links shared in chats and files them for you and your agents

Runs locallyStatus and history
Install
Pricing
Open source
Platforms
macOS, Windows, Linux
Runs
Locally
Version
1.2.1
Listing updated
today

Similar MCP servers

About driftnet

driftnet is an open-source MCP server for knowledge and memory. It runs locally on your machine and is started by your AI client.

A post that only names a project is a poor bookmark, so driftnet goes further than saving the URL. It follows a post to the repo it is really about, looks projects up on GitHub by exact name when a caption lists them without links, and discards links a model invents. Pages are cut to an outline of about 350 tokens before any model sees them, and the README reports five Facebook reels judged for $0.00029 in total.

Links arrive from a Chrome extension, Discord, Telegram, Slack or LINE bots, the command line, a single HTTP call or an agent over MCP. Each becomes a card with pictures, a one-line answer, what you will see and why it is there, filed into your own folders and kept in local SQLite with full-text search that handles Chinese. New items wait in an Inbox with suggested folders, and a mind map lets you drag cards into place.

It works with any model or none. Providers include Anthropic, OpenAI, Gemini, Bedrock, Vertex, the Claude and Gemini CLIs, or any OpenAI-compatible endpoint such as Ollama; without a model, links are saved unsorted under Other. It runs with npx driftnet on Node 22.13 or newer and serves a page on localhost:8787. Private and local addresses are refused.

Install

Run in a terminal

claude mcp add driftnet --env DRIFTNET_LLM_API_KEY='<DRIFTNET_LLM_API_KEY>' --env AWS_ACCESS_KEY_ID='<AWS_ACCESS_KEY_ID>' --env AWS_SECRET_ACCESS_KEY='<AWS_SECRET_ACCESS_KEY>' --env AWS_BEARER_TOKEN_BEDROCK='<AWS_BEARER_TOKEN_BEDROCK>' -- npx -y driftnet

Add to claude_desktop_config.json

{
  "mcpServers": {
    "driftnet": {
      "command": "npx",
      "args": [
        "-y",
        "driftnet"
      ],
      "env": {
        "DRIFTNET_LLM_API_KEY": "<DRIFTNET_LLM_API_KEY>",
        "AWS_ACCESS_KEY_ID": "<AWS_ACCESS_KEY_ID>",
        "AWS_SECRET_ACCESS_KEY": "<AWS_SECRET_ACCESS_KEY>",
        "AWS_BEARER_TOKEN_BEDROCK": "<AWS_BEARER_TOKEN_BEDROCK>"
      }
    }
  }
}

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "driftnet": {
      "command": "npx",
      "args": [
        "-y",
        "driftnet"
      ],
      "env": {
        "DRIFTNET_LLM_API_KEY": "<DRIFTNET_LLM_API_KEY>",
        "AWS_ACCESS_KEY_ID": "<AWS_ACCESS_KEY_ID>",
        "AWS_SECRET_ACCESS_KEY": "<AWS_SECRET_ACCESS_KEY>",
        "AWS_BEARER_TOKEN_BEDROCK": "<AWS_BEARER_TOKEN_BEDROCK>"
      }
    }
  }
}

Add to .vscode/mcp.json

{
  "servers": {
    "driftnet": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "driftnet"
      ],
      "env": {
        "DRIFTNET_LLM_API_KEY": "<DRIFTNET_LLM_API_KEY>",
        "AWS_ACCESS_KEY_ID": "<AWS_ACCESS_KEY_ID>",
        "AWS_SECRET_ACCESS_KEY": "<AWS_SECRET_ACCESS_KEY>",
        "AWS_BEARER_TOKEN_BEDROCK": "<AWS_BEARER_TOKEN_BEDROCK>"
      }
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "driftnet": {
      "command": "npx",
      "args": [
        "-y",
        "driftnet"
      ],
      "env": {
        "DRIFTNET_LLM_API_KEY": "<DRIFTNET_LLM_API_KEY>",
        "AWS_ACCESS_KEY_ID": "<AWS_ACCESS_KEY_ID>",
        "AWS_SECRET_ACCESS_KEY": "<AWS_SECRET_ACCESS_KEY>",
        "AWS_BEARER_TOKEN_BEDROCK": "<AWS_BEARER_TOKEN_BEDROCK>"
      }
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "driftnet": {
      "command": "npx",
      "args": [
        "-y",
        "driftnet"
      ],
      "env": {
        "DRIFTNET_LLM_API_KEY": "<DRIFTNET_LLM_API_KEY>",
        "AWS_ACCESS_KEY_ID": "<AWS_ACCESS_KEY_ID>",
        "AWS_SECRET_ACCESS_KEY": "<AWS_SECRET_ACCESS_KEY>",
        "AWS_BEARER_TOKEN_BEDROCK": "<AWS_BEARER_TOKEN_BEDROCK>"
      }
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

Configuration

  • DRIFTNET_HOMEWhere settings and the catalogue live (default ~/.driftnet)
  • DRIFTNET_LLM_PROVIDERanthropic, openai, gemini, bedrock, vertex, claude-cli or gemini-cli; empty for no summaries
  • DRIFTNET_LLM_MODELModel name for that provider
  • DRIFTNET_LLM_API_KEYAPI key for that provider (gemini also reads GEMINI_API_KEY; for vertex it may be a raw access token)
  • DRIFTNET_LLM_BASE_URLOpenAI-compatible endpoint for the openai provider (Ollama, OpenRouter, Gemini…)
  • DRIFTNET_LLM_REGIONRegion for the bedrock provider (or AWS_REGION) and the vertex provider (default us-central1)
  • AWS_ACCESS_KEY_IDbedrock provider: AWS access key ID (with AWS_SECRET_ACCESS_KEY, and AWS_SESSION_TOKEN for temporary credentials)
  • AWS_SECRET_ACCESS_KEYbedrock provider: AWS secret access key
  • AWS_BEARER_TOKEN_BEDROCKbedrock provider: a Bedrock API key, used instead of the AWS keys
  • GOOGLE_APPLICATION_CREDENTIALSvertex provider: path to a service-account JSON key file (or DRIFTNET_VERTEX_CREDENTIALS)
  • GOOGLE_CLOUD_PROJECTvertex provider: Google Cloud project id (optional when the key file names one)
  • DRIFTNET_JUDGEWho files each link before the model: jev, rules (free, no model), llm, heuristic, off, or a plugin's judge name
  • DRIFTNET_JUDGE_RULESPath to a rules file for the rules judge, one 'pattern -> folder' per line
  • DRIFTNET_LANGUAGELanguage summaries are written in (default English)
  • DRIFTNET_AGENTName recorded for this agent's saves and votes (default local)

What it does

Finds the real tool
Follows a post to the repo it covers, and drops links a model makes up.
Cards with summaries
Each link gets a one-line answer, what you will see and why it is saved.
Folder-based filing
Uses your folder paths and descriptions; suggestions wait in an Inbox.
Mind map sorting
Drag cards onto folders; AI tidy-up proposes changes that apply only on approval.
Many ways in
Chrome button, chat bots, CLI, HTTP call or MCP.
Any model or none
Supports several providers, local models via Ollama, or no model at all.

Good for

  • Collect links dropped in a team chat into one searchable library
  • Turn a weekly AI roundup post into one card per project
  • Let a coding agent read and add to your saved research over MCP
  • Import old bookmarks and sort them into folders

Pricing

Open source

Runs locally; any model API costs are your own, or use a free local model

driftnet FAQ

Do I need an account or server?

No. It runs on your own computer and stores data in local SQLite at ~/.driftnet.

Does it require an AI model?

No. Without one, links are saved and wait unsorted under Other, and an agent can write the summaries.

Which chat platforms are supported?

Discord, Telegram, Slack and LINE bots, plus a Chrome extension.

How do I start it?

Run `npx driftnet` with Node.js 22.13 or newer; it opens localhost:8787.

Is driftnet free?

Yes. driftnet is open source, so you can use and self-host it for free.

How do I add driftnet to Claude Code?

Run this command in your terminal: claude mcp add driftnet --env DRIFTNET_LLM_API_KEY='<DRIFTNET_LLM_API_KEY>' --env AWS_ACCESS_KEY_ID='<AWS_ACCESS_KEY_ID>' --env AWS_SECRET_ACCESS_KEY='<AWS_SECRET_ACCESS_KEY>' --env AWS_BEARER_TOKEN_BEDROCK='<AWS_BEARER_TOKEN_BEDROCK>' -- npx -y driftnet

What are the best alternatives to driftnet?

The closest alternatives to driftnet are Codebase Memory, XERJ, Ha MCP, Claude Code Ultimate Guide and Concord MCP.

Where this comes from

  • Description: written from the project's README.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers