
Vscode Terminal Mcp
io.github.sirlordtPublished by GitHub account sirlordtv0.1.0Updated Oct 24, 2026
Execute commands in visible VSCode terminal tabs with output capture and session reuse.
The publisher's own server: Terminal Tax
Overview
An MCP server that runs commands in visible VS Code terminal tabs, capturing output and reusing sessions so an assistant can execute and monitor shell commands.
- What you can do
- Run a command in a new or reused visible terminal and get clean output with exit code and timing, or launch long-running commands without waiting and poll their output incrementally with tail mode. Create, list, filter by agent tag, and close terminal sessions, including isolated terminals for parallel subagents. Send text into interactive terminals to answer prompts, REPLs, or confirmations. Read buffered output with pagination; oversized output is saved to a temporary JSON file whose path is returned.
- When to use it
- Useful when you want an assistant to run builds, tests, installs, or deployments where the user should watch progress live in VS Code, or to drive interactive command-line prompts and keep persistent shell sessions across steps.
- Requirements
- Local process started via npx from the npm package vscode-terminal-mcp; needs Node.js 20+ and VS Code 1.93+ with the companion extension installed, since commands run through the VS Code Shell Integration API over a local IPC socket. No accounts, API keys, or environment variables are declared.
Installation
In SourceWeft
- Open Vscode Terminal Mcp in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
vscode-terminal-mcp
MCP server that executes commands in visible VSCode terminal tabs with full output capture. Unlike inline execution, every command runs in a real terminal you can see, scroll, and interact with.
Key Features
- Visible Terminals: Commands run in real VSCode terminal tabs, not hidden processes. You see everything in real time.
- Session Reuse: The
runtool automatically reuses idle sessions, creating new terminals only when needed. - Long-Running Support: Fire-and-forget execution with
waitForCompletion: false, then poll output incrementally withread. - Subagent Isolation: Tag sessions with
agentIdto keep parallel agent workloads separated.
Requirements
- VS Code 1.93+ (for Shell Integration API)
- Node.js 20+
Getting Started
Claude Code
VS Code / Copilot
Add to your .vscode/mcp.json:
Cursor
Add to your .cursor/mcp.json:
Claude Desktop
Add to your claude_desktop_config.json:
Your First Prompt
After installation, try asking:
Run
ls -lain the terminal
You should see a new terminal tab open in VSCode with the command output.
Screenshots
Running a command with run
Permission dialog for exec
Exec result with clean output
Tools
Quick Execution
Session Management
Usage Patterns
Simple Command
The run tool handles everything — creates a terminal if needed, executes, and returns clean output:
Long-Running Process
For builds, deployments, or any command that takes a while:
The agent will:
- Call
runwithwaitForCompletion: false— returns immediately - Call
readwithoffset: -10to check the last 10 lines - Repeat until the process completes
Interactive Commands
For commands that need user input:
The agent will:
- Call
runwithnpm init - Call
readto see the prompt - Call
inputto send the answer
Parallel Agents
Subagents can work in isolated terminals using agentId:
Each subagent gets its own terminal tagged with its agentId, preventing output from mixing.
Configuration
The extension reads configuration from VSCode settings under terminalMcp.*:
Recommended: Set as Preferred Tool
LLM agents like Claude Code have a built-in Bash tool that executes commands inline within the chat. The output is embedded in the conversation and hard to read, especially for commands with verbose output. We recommend telling the agent to prefer this MCP over its built-in Bash tool.
Add the following to your project's CLAUDE.md (or equivalent instructions file):
Why this matters:
Development: Updating the Extension
VSCode aggressively caches extensions in memory. When developing locally, code --install-extension and even "Developer: Reload Window" may not reload your changes. Use this workflow:
Quick update (no restart needed)
After modifying source files, build and copy directly into the installed extension directory:
Then run "Developer: Reload Window" (Ctrl+Shift+P).
Full reinstall (when quick update doesn't work)
If VSCode still uses old code:
Verify the correct version is loaded
Large Output Handling
When read returns output that exceeds the MCP client's token limit, the system automatically saves the full output to a temporary JSON file and returns the file path in the error message.
To extract the relevant content:
The file format is JSON: [{"type": "text", "text": "..."}]
This commonly happens with commands that produce heavy TUI output (progress bars, ANSI escape codes). Use smaller offset values (e.g., offset: -20 instead of offset: -100) to reduce the captured output size.
How It Works
- The VSCode extension activates and starts an IPC server on a Unix socket
- The MCP entry point (
mcp-entry.js) is spawned by the MCP client and bridges JSON-RPC stdio with the IPC socket - Commands execute in real VSCode terminals using the Shell Integration API for reliable output capture and exit code detection
- Output is stored in circular buffers with pagination support for efficient reading
Latest Changes (0.1.6)
- Screenshots in README for marketplace
- Clean output format for all tools — no more raw JSON
- Fixed
waitForCompletion: falsenot working - Disabled idle reaper — user closes sessions manually
- Unique IPC socket per workspace (multi-instance support)
- Custom terminal tab names with date format
- Large output handling documentation
See CHANGELOG.md for full history.
License
MIT
Source: README.md at commit 60df01d
Tools
0Version history
1- v0.1.0LatestSep 16, 2026
Score
50% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts
Other Terminal integrations

Terminal Tax
tax.terminalOfficial publisher
Hosted tax-filing server that lets an assistant send a US federal 2025 individual return's facts, get it computed and checked, and hand back a private signing…
Terminal
mn.dtaOfficial publisher
A remote MCP server for Mongolia's signed macro and market data archive, letting an assistant search series, pull panels and verify signed observations.
SCRIPTMASTERLABS — Quant Terminal
com.scriptmasterlabs
Remote MCP endpoint offering real-data ticker levels and a "squeeze-mirror" projection for quantitative trading analysis.

Terminal 7
Published by GitHub account LinkMarine777
Signals, scores and payout data for stonkfun coins on Solana, for bots and AI agents.
terminal-use
Published by GitHub account computer-agent-labs
Lets an AI assistant drive real terminal sessions on your machine: type, press keys, click, read the screen and take screenshots, including full-screen TUIs.