
Ag Bash
io.github.sairam0424Published by GitHub account sairam0424v6.0.4Updated Oct 24, 2026
Sandboxed AI-native bash with 70 agentic tools: run_bash, code edit/diff, WASM Python/JS.
Overview
A local MCP server that gives an AI assistant a persistent, sandboxed Bash environment with around 70 tools for running scripts and editing code.
- What you can do
- Run shell scripts with a full Unix toolkit such as jq, grep, sed and awk, and keep the environment persistent across calls (run_bash). Inspect the current working directory, environment variables and defined functions (get_state). Edit and diff code, find symbols, explain and analyze code, and take snapshots through the agentic suite (ag-edit, ag-diff, ag-hover, ag-explain, ag-find-symbol, ag-analyze, ag-snapshot). Plan work, keep notebooks and todos, and run Python or JavaScript through WASM.
- When to use it
- Useful when an assistant needs to execute shell commands, inspect a project's state, or make and review code changes inside a controlled sandbox rather than on the raw host. Suited to coding and scripting workflows in Claude Desktop, Claude Code, Cursor or VS Code.
- Requirements
- Runs as a local stdio process on the user's machine (desktop only). Needs Node.js to run the npm package @ag-bash/mcp-server, for example via npx or a global install; Homebrew is also mentioned. No authentication, environment variables or headers are declared.
Installation
In SourceWeft
- Open Ag Bash 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
@ag-bash/mcp-server
MCP server exposing 70 ag-bash tools to Claude Desktop and other MCP hosts
[npm version] [MCP Registry] [License: Apache-2.0]
A standalone Model Context Protocol (MCP) server that gives AI agents a persistent, sandboxed Bash environment, exposing 70 tools over stdio.
Installation
Add the server to Claude Code in one line:
Or install it globally:
Claude Code plugin
Install via the self-hosted marketplace:
MCP Registry
Listed as io.github.sairam0424/ag-bash (active) at registry.modelcontextprotocol.io.
Also available via: Homebrew (ag-bash-mcp bin) · Docker MCP Catalog (in review)
Usage
Run the server directly:
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
Cursor / VS Code
Add a new MCP server in your IDE settings:
- Name:
ag-bash - Type:
command - Command:
ag-bash-mcp
What's Exposed
run_bash— Execute scripts with full unix toolkit (jq,grep,sed,awk, etc.)get_state— Inspect CWD, environment variables, and defined functions- Agentic suite —
ag-edit,ag-diff,ag-hover,ag-explain,ag-find-symbol,ag-todo,ag-analyze,ag-snapshot,ag-plan,ag-notebook,ag-mcp - 70 tools total via JSON-RPC 2.0 with Zod schema validation on every input
Features
- Rate limiting — 60 requests/minute per session (configurable)
- Read-only root — Project files mounted as overlay; writes stay in memory
- Sanitized errors — JSON-RPC responses strip file paths, capped at 200 chars
- Zero console leakage — All diagnostics flow through structured handlers
- Resource limits — Protection against infinite loops and excessive memory
- Orchestration governance — Agent nesting limits prevent recursive loops
v4.1 Upgrade Notes
Links
- GitHub Repository
- Core Engine
- MCP Registry —
io.github.sairam0424/ag-bash
License
Apache-2.0
Source: packages/mcp-server/README.md at commit efac5c3
Tools
0Version history
1- v6.0.4LatestSep 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