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

Screenpipe

Official
Search local screen recordings, audio transcripts, and computer activity.
How to install ↓ Compare GitHub source repository ↗
Category
Other #150 of 230
Stars
★ 22.0k Very popular
Transport
stdio (local process)
Runtime
Node.js
Credentials
No credential needed
License
Other / unspecified
Last commit
50FMRS · D

Provided in the screenpipe repository, this version 0.19.0 npm package uses stdio transport. It is suited to locally searching screen and audio history and supplying work context to AI assistants, but users should evaluate sensitive-data capture, system permissions, resource usage, and the data boundaries of optional cloud features.

Strongest · Documentation 14/20 Weakest · Reliability 5/20

Reliability
5/20
Security and permissions
8/20
Maintenance
11/20
Documentation
14/20
Setup experience
12/20
Why each score
Reliability 5/20
The manifest and README do not verify that the server starts, completes the MCP initialization and tool-list handshake, or implements the declared behavior. No committed tests or CI evidence for the MCP key paths is provided. The README makes a plausible stdio installation path, but tool behavior, dependency control, error handling, and edge cases are unverified, so the score is heavily reduced.
Security and permissions 8/20
Local storage, optional local models, and privacy filters are positive evidence. However, the README also discloses default PostHog/Sentry telemetry, optional cloud transcription and sync, a raw SQL API, and pipes that can call APIs, write files, and take actions. There is no source evidence of tool-level authorization, authentication, dangerous-operation confirmation, or complete data-flow controls. No explicit evidence meets a red-line condition.
Maintenance 11/20
The repository is not archived and provides version 0.19.0, maintainer information, and contribution guidance, indicating ongoing activity. However, no commit history, release cadence, dependency-update evidence, vulnerability-response channel, or demonstrated issue response is supplied. The license metadata is NOASSERTION while the README claims a commercial license, creating governance and license-audit gaps.
Documentation 14/20
The README covers installation, client examples, platforms, resource usage, privacy, telemetry, cloud processing, configuration, and several limitations, making it relatively usable. Deductions apply because the README is untrusted evidence and does not provide MCP tool names, parameter schemas, return formats, error codes, compatibility guarantees, or systematic troubleshooting. Its license statement also conflicts with the supplied metadata.
Setup experience 12/20
A short npx installation path, a Claude MCP configuration example, stdio transport, and desktop/CLI options are documented for mainstream use. The score is limited because prerequisites, background-service requirements, platform differences, version pinning, startup diagnostics, and configuration for other clients are not verified from source files, and the connection was not executed.

Test out of dateListed 2026-07-29

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesConnects to a database

Best for

  • Knowledge workers who need to retrieve personal screen and audio history.
  • Users who prioritize local processing and privacy.
  • Developers who want to provide work context to Claude, Cursor, or other MCP clients.
  • Remote workers who need meeting transcription and activity review.

Not for

  • Users who do not want continuous screen or audio capture.
  • Teams that require centralized cloud storage without local recording.
  • Use cases requiring only ordinary file-system or database access.

Required permissions

  • Screenpipe needs applicable screen-recording permissions to capture screen content.
  • System-audio and microphone capture require the relevant system audio or microphone permissions.
  • The application uses the accessibility tree to extract screen text and falls back to OCR when unavailable.
  • On macOS, system-audio capture requires Screen & System Audio Recording permission and the relevant experimental feature to be enabled.

Risks and side effects

  • Screens, audio, transcripts, and search indexes may contain passwords, private conversations, or proprietary information.
  • Data is stored locally by default, but enabled analytics and crash diagnostics may send device or error metadata.
  • Cloud transcription, hosted AI, or cloud sync can remotely process related audio, prompts, context, or synchronized data.
  • Continuous capture consumes system resources; the README reports typical CPU usage of 5–10%.

Setup

Before you start

Runtime：Node.js

Other optional settings (1)
SCREENPIPE_AUDIO_EXCLUSIONS_PATH optional Optional; overrides the path to the system-audio exclusion list file (default ~/.screenpipe/audio-exclusions.), mainly for testing.

Run npx screenpipe record, then run npx screenpipe setup. With Claude, you can also run claude mcp add screenpipe -- npx -y screenpipe-mcp@latest. Then query recent screen activity or ask for a summary in an MCP-compatible client.

Check that it works

Run `claude mcp add screenpipe -- npx -y screenpipe-mcp@latest`, then ask your client 'What did I see in the last 5 mins?' — a grounded answer from local screen history confirms the server is connected.

Troubleshooting

  1. Confirm that npx screenpipe record is running and npx screenpipe setup has been completed.
  2. Confirm that the MCP client starts the server with npx -y screenpipe-mcp@latest.
  3. If searches return nothing, verify that screenpipe captured the relevant screen or audio content and that recording permissions are granted.
  4. If system audio is unavailable, check macOS permissions, the experimental CoreAudio system-audio setting, and audio-exclusion configuration.
  5. For local-only processing, keep cloud sync disabled and select local transcription and a local AI provider.

Things to try

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

  • What did I see in the last 5 mins?
  • Summarize today's conversations
  • Search my screen history for meeting notes
  • What was said in my most recent meeting?

Use cases

Search recent screen content using natural language.
Query audio transcripts and meeting conversations.
Give AI assistants context from local work history.
Create automations based on computer activity.

Supported clients

Claude Desktop
Cursor
VS Code
Cline
Continue

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

Overview

The screenpipe MCP server lets compatible AI assistants query local screen recordings, audio transcripts, and computer activity stored by screenpipe. screenpipe stores data locally by default and supports natural-language search, timeline review, and meeting-transcript queries.

Similar servers

BioMCP 75 · B

One binary. One grammar. Evidence from the biomedical sources you already trust.

★ 659 · Tools 11 Compare with this →
Compare with other MCP servers

Source revision 62482a789d03 Data synced 2026-10-19 Read the FMRS scoring method