we launched on YC — see the post →
Claude logo

screenpipe × Claude

Claude Desktop + Claude Code

Give Claude the work context you already have

screenpipe captures the screen, audio, and meeting context where work happens. Claude can retrieve the relevant slice through MCP when a chat needs more context.

Checked against Screenpipe's current connection code and documented setup on July 24, 2026.

The useful difference

The repository or chat is only part of the task.

screenpipe retrieves context from the apps, meetings, browser, and terminal around it. The agent gets the selected result, not an unbounded dump of your history.

What this unlocks

Useful context for Claude, on demand

Recall what was decided

Search meeting transcripts and surrounding screen context instead of relying on a summary copied into the chat.

Resume interrupted research

Find the docs, pages, terminal output, and conversations you had open before the current Claude session started.

Retrieve exact evidence

Ask for a specific phrase, error, app, time range, or conversation and inspect the source context Claude used.

Turn a workflow into a Pipe

Use captured examples of repeated work to draft a scheduled Screenpipe Pipe with a human review step.

Setup

Connect screenpipe to Claude

Use the app path for the safest default. The manual path is there when you want to inspect or own the configuration directly.

Recommended for Claude Desktop

Connect from the screenpipe app

The app uses its bundled runtime, writes the Claude Desktop MCP entry, and includes the current local API key.

  1. 1Open screenpipe and confirm recording is running.
  2. 2Go to Settings, Connections, Claude Desktop, then select Connect.
  3. 3Restart Claude Desktop and ask what you did in the last five minutes.

verify: Open Claude's MCP tools and confirm screenpipe is connected.

For Claude Code

Add the local MCP server from your terminal

This uses Claude Code's supported stdio MCP setup at user scope, so the connection is available across your local projects.

  1. 1Run the command below from a terminal with Node.js 18 or newer.
  2. 2Start a new Claude Code session after the command finishes.
  3. 3Run /mcp or claude mcp list if you want to verify the connection.
claude mcp add --transport stdio --scope user screenpipe -- npx -y screenpipe-mcp@latest

verify: claude mcp list

Connect every detected AI tool

Use the screenpipe setup command

The Screenpipe CLI detects supported tools on the computer, preserves unrelated configuration, and installs the available MCP and skill integrations.

  1. 1Run the command once.
  2. 2Review the tools Screenpipe detected and connected.
  3. 3Restart the AI tools you plan to use.
npx -y screenpipe@latest setup

Data boundary

Local capture does not mean the agent sees nothing

MCP is useful because selected results enter the agent's context. Treat that handoff as an explicit data boundary and choose what you capture, search, and share.

Capture and index

In local-only mode, raw screen and audio capture plus the searchable index stay on the device.

MCP retrieval

The local server searches for the requested phrase, app, meeting, memory, or time range and returns the selected result.

Claude context

The selected result is sent into Claude's context and is then handled under that provider's settings and policies.

Cloud AI, sync, connectors, exports, and team workflows are separate choices. Review the privacy and data-flow guide before using sensitive work context with an external model.

Real workflows

Prompts that use context Claude did not have

Recover a meeting decision

Bring the discussion and the screen context around it into the current chat.

Search my screen and audio from yesterday's product meeting. What did we decide about onboarding, and what evidence supported it?

Continue research without starting over

Find the sources and notes you were using before the current conversation.

Find the documentation and browser research I reviewed this morning about MCP permissions. Summarize the relevant passages and include the source times.

Locate an exact conversation

Search audio and screen text together when a name or phrase is only partly remembered.

When did someone mention a seven-day trial this week? Find the surrounding conversation and summarize the concern.

Draft a repeatable automation

Use real examples of the workflow before asking Claude to automate it.

Find three times I turned customer feedback into an issue. Compare the steps, then draft a Screenpipe Pipe prompt with a review checkpoint before creating anything.

Troubleshooting

Common connection issues

Claude Desktop does not show screenpipe

Confirm screenpipe is running, reconnect from Settings, Connections, Claude Desktop, then restart Claude Desktop. The desktop and Claude Code setup paths are separate.

Claude Code cannot start the MCP server

The manual Claude Code path requires Node.js 18 or newer. The app's bundled-runtime connection applies to Claude Desktop, not Claude Code.

A search returns no useful result

Check that screenpipe was recording during the requested time, then narrow the request by app, phrase, or time range. MCP can only retrieve context that was captured and indexed.

Keep exploring

Give Claude the missing context

Start screenpipe, connect the agent, and test one concrete recall question from work you did today.