MIT-licensed, Python control plane that turns DeepSeek Harness into a scoped shell/filesystem/remote-execution agent via a DSH plugin bundle; container/VM-bounded safety, OAuth, audit logs, and remote-worker reach into firewalled machines.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-09-02
License
MIT
01What can it help you accomplish?
Run shell, filesystem, Git, and Playwright tasks from DeepSeek Harness against a controlled workspace
A scoped /workspace with shell, persistent shell, filesystem, search, patch, Playwright, audit, durable logical sessions, file links, and outbound remote-worker access
DeepSeek Harness users who want to give an LLM a real, scoped execution environment instead of a sandboxed toy
Control NAT-, firewall-, or HPC-bound machines that can only connect outward, from a DSH conversation
One-time remote-worker invites and a `machine` argument on normal tools, so DSH can drive lab/HPC/NPU hosts through outbound HTTPS
Platform and infra teams running DeepSeek Harness against machines behind strict firewalls or without inbound connectivity
02How to install into DeepSeek Harness
Prerequisites
- A normal LSM HTTP/MCP service already running on the same host (the bundle installs into an existing service)
- Python 3.11+ (the README shows a python-3.11+ badge for the project)
- Docker, for the `docker compose up -d` runtime used in the quick start
- A DeepSeek Harness profile to install into (the README uses the `web` profile)
- OAuth values in `.env` for public/HTTPS deployments: `LOCAL_SHELL_MCP_AUTH_MODE=oauth`, an admin PIN, and a JWT secret
Installation steps
- 01
Start the LSM service on the host: `mkdir -p workspaces/default` then `docker compose up -d`, and confirm with `curl -i http://127.0.0.1:8765/healthz`
$ curl -i http://127.0.0.1:8765/healthz
- 02
Install the plugin bundle into a DSH profile: `dsh plugin --profile web add 'github:fwerkor/local-shell-mcp#main'`
$ dsh plugin --profile web add 'github:fwerkor/local-shell-mcp#main'
- 03
For production, pin the Git spec to a reviewed release or commit instead of the moving `main` branch
Verify the integration
- The health check returns 200: `curl -i http://127.0.0.1:8765/healthz`
- DSH exposes the model tools under the `mcp__lsm__*` namespace
03DSH integration and capability boundaries
Install the repo as a DSH plugin bundle via the dsh CLI; an LSM-aware Streamable HTTP bridge exposes the full LSM tool surface (including remote workers) inside your DSH profile.
Workspace shell & filesystem
shell commands, file paths, and patches under the controlled root→shell / persistent shell output, file read/write/edit/patch/grep/glob, and audit-tracked activity
runs commands inside the container/VMwrites and patches files under the workspace rootRemote workers
a one-time invite and an optional `machine` argument on normal tools→control of NAT/firewall/HPC/NPU/lab machines that connect outward over HTTPS
runs commands on remote worker machinesopens outbound HTTPS connections to remote endpointsBrowser automation (Playwright)
browser session actions and optional Playwright scripts→page text, PNG/PDF evidence captures, and full Playwright script runs
launches a headless browser and runs page scriptsDynamic MCP tools
external MCP server configs→progressively discovered external tools via `mcp_manage` / `mcp_tool_search` / `mcp_tool_inspect` / `mcp_tool_call`
spawns dynamic stdio MCP servers with a minimal OS environment
04Who is it for? When not to use it?
Good for
- DeepSeek Harness users who want to give an LLM a real, scoped execution environment instead of a sandboxed toy
- Platform and infra teams running DeepSeek Harness against machines behind strict firewalls or without inbound connectivity
Not for
- The connected model is treated as having control of the container or VM — the intended safety boundary is the container/VM, not the host. Run it in a disposable container or VM and do not mount the host root or /var/run/docker.sock.
- Never expose the service with `LOCAL_SHELL_MCP_AUTH_MODE=none` on a public network; use `oauth` and keep long-lived credentials out of model-visible environment variables.
05Compatibility, maintenance and safety notes
- The connected model is treated as having control of the container or VM — the intended safety boundary is the container/VM, not the host. Run it in a disposable container or VM and do not mount the host root or /var/run/docker.sock.
- Never expose the service with `LOCAL_SHELL_MCP_AUTH_MODE=none` on a public network; use `oauth` and keep long-lived credentials out of model-visible environment variables.
MIT · actively maintained (latest release v4.2.1, 2026-08-25)
06Frequently asked questions
What is the DeepSeek Harness integration mode?
The repository is a DSH plugin bundle. With a normal LSM service running on the same host, run `dsh plugin --profile web add 'github:fwerkor/local-shell-mcp#main'` to install it into a DSH profile; an LSM-aware Streamable HTTP bridge exposes the full LSM tool surface.
How are my DSH conversations isolated from each other?
Each DSH Session gets a stable v4 logical-session identity, so its Logical Session, active run, Activity, and native Live Workspace view stay isolated from other DSH conversations and survive DSH-side MCP transport recreation.
What tool namespace does DSH see?
DSH sees the model tools under the normal `mcp__lsm__*` namespace, including remote workers, browser tools, and Dynamic MCP tools.
Should I pin the version when installing?
For production the README advises pinning the Git spec to a reviewed release or commit rather than the moving `main` branch.
Is the service safe to expose publicly?
The intended safety boundary is the container or VM, not the host. Use OAuth, run it in a disposable container/VM, never expose it with `LOCAL_SHELL_MCP_AUTH_MODE=none` on a public network, and keep long-lived credentials out of model-visible env vars.
07Related DSH workflows
archify
by tt-a1i
Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
openviking
by volcengine
Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
nocobase
by nocobase
NocoBase is an open-source AI + no-code platform for building business systems fast. Instead of generating everything from scratch, AI works on top of production-proven infrastructure and a WYSIWYG no-code interface, so you get both speed and reliability.
learn-harness-engineering
by walkinglabs
Harness engineering beginner tutorial, from 0 to 1
08Data and sources
The repository root is also a DSH plugin bundle.
dsh plugin --profile web add 'github:fwerkor/local-shell-mcp#main'
The bundle uses an LSM-aware Streamable HTTP bridge and keeps the complete LSM tool surface, including `remote_manage`,…
DSH sees model tools under the normal `mcp__lsm__*` namespace.
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-09-02. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.