Local MCP server mapping MITRE ATT&CK techniques to NIST 800-53r5 controls to DISA STIG fix and check steps, severity-ordered
- ✓Open-source license (MIT)
- ✓Actively maintained (<30d)
- ✓Clear description
- ✓Documented (README)
claude mcp add stig-mcp -- uvx stig-mcp{
"mcpServers": {
"stig-mcp": {
"command": "uvx",
"args": ["stig-mcp"]
}
}
}MCP Servers overview
# stig-mcp
Local MCP server that maps MITRE ATT&CK® techniques (and actors) to the NIST
800-53r5 controls that mitigate them, with the DISA STIG fix and check steps for
the systems under consideration, severity-ordered.
<!-- mcp-name: io.github.jeneric/stig-mcp -->
## Install
stig-mcp is on [PyPI](https://pypi.org/project/stig-mcp/). With
[uv](https://docs.astral.sh/uv/), a client runs it with no separate install step:
uvx stig-mcp
`uvx` comes with uv; [install uv](https://docs.astral.sh/uv/getting-started/installation/)
first if `uvx --version` does not run. The VS Code badge and the Claude Code plugin below both
need it.
From a source checkout instead, install the dependencies with:
uv sync
The commands below are written for a checkout (`uv run ...`). Without one, run the same entry
point with `uvx --from stig-mcp`, for example `uvx --from stig-mcp stig-mcp-install-kb`.
## Quickstart
1. Wire the server into a client (see "Run the server" below) and start it.
2. Ask the agent to install the knowledge base. It calls the `install_knowledge_base`
tool, which downloads the newest published release from this project's GitHub releases
and verifies its SHA-256 before installing it. From a terminal the same is
`uv run stig-mcp-install-kb`. A host that cannot reach GitHub installs from a file; see
[docs/operations.md](https://github.com/jeneric/STIG-MCP/blob/main/docs/operations.md), "Install a prebuilt knowledge base".
3. Or build it yourself: `uv run stig-mcp-fetch` downloads ATT&CK, the CTID mapping, the
800-53 catalog, and DISA's STIG content. **This transfers roughly a gigabyte** and
refuses to start with less than 2 GiB free. Then `uv run stig-mcp-ingest` builds the
knowledge base.
On a host that cannot reach `dl.dod.cyber.mil`, place the artifacts in the sources
directory yourself and go straight to `stig-mcp-ingest`. That is a first-class path rather
than a fallback: the ingest reads a directory and never consults the fetch. See
[docs/operations.md](https://github.com/jeneric/STIG-MCP/blob/main/docs/operations.md), "Placing the sources by hand".
Afterwards, `uv run stig-mcp-fetch --check` reports what MITRE ATT&CK, CTID, NIST and DISA
have published since, exiting 10 when there is something to take and 3 when a source could not
be reached, and `--refresh` takes it.
Neither rebuilds the knowledge base; see "Keeping current" in the same document.
## Run the server
uvx stig-mcp
or, from a checkout, `uv run stig-mcp`.
This is a stdio MCP server: it speaks JSON-RPC on stdin/stdout and logs to stderr,
so it is launched by an MCP client rather than run standalone.
It starts whether or not the knowledge base exists, and it never answers from one it
cannot trust. Called before the knowledge base is installed, or against one an older release
wrote, every tool returns a `not_ready` payload instead of an answer: the reason, which
source files it can and cannot see, the sources directory it looked in, and the next steps,
led by the `install_knowledge_base` tool and followed by the commands to run. Each command
comes in two forms: `run`, which works from the server's own environment, and `as_installed`,
which a person can type into a terminal, written for how the server was installed (`uvx`, a
checkout, or an installed copy). That is deliberate, so an agent can read the remedy from
the tool result rather than the operator having to find a log pane. Install or rebuild the
knowledge base and the running server picks it up without a restart.
The `check_sources` tool tells an agent whether a newer knowledge base is published. It and
`install_knowledge_base` are the only two tools that contact the network, and they reach
only this project's GitHub releases.
### GitHub Copilot in VS Code
[](https://vscode.dev/redirect/mcp/install?name=stig-mcp&config=%7B%22type%22%3A%22stdio%22%2C%22command%22%3A%22uvx%22%2C%22args%22%3A%5B%22stig-mcp%22%5D%7D)
Or run **MCP: Open User Configuration** from the Command Palette and add:
```json
{
"servers": {
"stig-mcp": {
"type": "stdio",
"command": "uvx",
"args": ["stig-mcp"]
}
}
}
```
From a checkout, create `.vscode/mcp.json` in this repository instead (git-ignored, so it
stays local):
```json
{
"servers": {
"stig-mcp": {
"type": "stdio",
"command": "uv",
"args": ["run", "stig-mcp"],
"cwd": "${workspaceFolder}"
}
}
}
```
Then:
1. Open Copilot Chat and set the mode dropdown to **Agent**. MCP tools are not
available in Ask or Edit mode.
2. Command Palette (`Ctrl+Shift+P`, or `Cmd+Shift+P` on macOS) and run
**MCP: List Servers**, select `stig-mcp`, then **Start**. Trust the server when
prompted, since it runs a local command.
3. Click **Configure Tools** in the chat input to confirm the eight tools are listed
and enabled.
4. Reference a tool explicitly to verify the wiring, rather than hoping the model
picks it up on its own. See [docs/user-guide.md](https://github.com/jeneric/STIG-MCP/blob/main/docs/user-guide.md)'s "Getting the
LLM to use the server" for a prompt shape that reliably does this.
Copilot saves a tool answer over 8 KB to a temporary file and reads it back, so with
manual permissions it asks to read a file named like `…copilot-tool-output-….txt`
outside the workspace. That file is this server's answer; allow it.
To debug, run **MCP: List Servers**, select the server, and choose **Show Output**.
The two common failures are that `uvx` or `uv` is not on the `PATH` VS Code inherited,
which looks like a broken server but is a missing command, and an absent knowledge base.
For the first, use the absolute path (`which uvx` or `which uv`) as `command`. A missing
`uvx` shows in VS Code's output as `Connection state: Error spawn uvx ENOENT`, and in
`claude mcp list` as `Failed to connect — ENOENT: Executable not found in $PATH: "uvx"`. For
the second, see [docs/operations.md](https://github.com/jeneric/STIG-MCP/blob/main/docs/operations.md).
### Other clients
Any MCP client that launches a stdio server works, with `uvx stig-mcp` as the command. For
Claude Code, install the plugin from this repository's marketplace, inside a session (Claude
Code 2.1.275 or later):
/plugin install stig-mcp --marketplace jeneric/STIG-MCP
or from a shell, `claude plugin marketplace add jeneric/STIG-MCP` then
`claude plugin install stig-mcp@stig-mcp`. The plugin pins the current release, and
`claude plugin update stig-mcp@stig-mcp` moves it to the next one. Without the plugin:
claude mcp add stig-mcp -- uvx stig-mcp
From a checkout, `uv run` locates the project from the working directory, so a client that
starts elsewhere needs `--directory`, which makes the command independent of where it is
launched:
uv run --directory /path/to/STIG-MCP stig-mcp
or, from the repository root, `claude mcp add stig-mcp -- uv run stig-mcp`.
By default the knowledge base is not found relative to the working directory, so only
`uv run` cares where the client starts the server. Where it *is* found depends on whether this is a
checkout or an installed copy. (A relative `STIG_MCP_DATA` does resolve against the working
directory, so give it an absolute path if the client's is not yours.)
### Where the data lives
Two environment variables override the defaults, and the defaults differ between a source
checkout and an installed copy (which includes `uvx stig-mcp`):
| | source checkout | installed, POSIX and macOS | installed, Windows |
|---|---|---|---|
| data directory (`STIG_MCP_DATA`) | `stig_mcp/data/` | `$XDG_DATA_HOME/stig-mcp`, else `~/.local/share/stig-mcp` | `$XDG_DATA_HOME/stig-mcp`, else `%LOCALAPPDATA%\stig-mcp`, else `~\.local\share\stig-mcp` |
| mapping overrides (`STIG_MCP_OVERRIDES`) | `overrides.yaml` at the repository root | `$XDG_CONFIG_HOME/stig-mcp/overrides.yaml`, else `~/.config/stig-mcp/overrides.yaml` | `$XDG_CONFIG_HOME/stig-mcp/overrides.yaml`, else `%LOCALAPPDATA%\stig-mcp\overrides.yaml`, else `~\.config\stig-mcp\overrides.yaml` |
A checkout is a directory holding both the package and the `pyproject.toml` that declares
it, so an editable install counts as one. XDG is used on POSIX, including macOS. On Windows
with the XDG variables unset, the default is `%LOCALAPPDATA%`, because a roaming profile
copies `~/.local/share` at every logon and logoff, and this project's downloads can run to a
gigabyte; when
`%LOCALAPPDATA%` is set this puts the mapping overrides file inside the data directory rather
than beside it, since Windows has one such variable rather than XDG's separate data and config
locations. (With `%LOCALAPPDATA%` unset, Windows falls back to the same separate `~/.config`
and `~/.local/share` trees POSIX uses, so the two stay apart in that case, same as the table
above shows.)
**The XDG variables are read first on every platform, Windows included**, as the table's
Windows column shows: a Windows host with `XDG_DATA_HOME` set uses it and never reaches
`%LOCALAPPDATA%`, so the roaming argument above holds only where that variable is unset. The
order is kept so that an existing install's data directory never moves under it.
`STIG_MCP_DATA` and `STIG_MCP_OVERRIDES` outrank everything above and are the escape hatch
everywhere, for a native location or any other.
`stig-mcp-ingest` creates the data directory if it does not exist. It refuses to run when
`STIG_MCP_OVERRIDES` names a file that is not there, rather than silently applying no
overrides; a missing file at the default location is fine, because that file is optional.
## What this server fetches
- The MCP server contacts nothing unless `check_sources` or `install_knowledge_base` is
called. Then it sends HTTPS GET requests to `api.github.com` (this repository's release
listing) and `github.com` (`/jeneric/STIG-MCP/releases/download/...`), which redirects to
`rWhat people ask about STIG-MCP
What is jeneric/STIG-MCP?
+
jeneric/STIG-MCP is mcp servers for the Claude AI ecosystem. Local MCP server mapping MITRE ATT&CK techniques to NIST 800-53r5 controls to DISA STIG fix and check steps, severity-ordered It has 0 GitHub stars and its last recorded update is dated 2026-10-01.
How do I install STIG-MCP?
+
You can install STIG-MCP by cloning the repository (https://github.com/jeneric/STIG-MCP) or following the README instructions on GitHub. ClaudeWave also provides quick install blocks on this page.
Is jeneric/STIG-MCP safe to use?
+
Our security agent has analyzed jeneric/STIG-MCP and assigned a Trust Score of 87/100 (tier: Trusted). See the full breakdown of passed checks and flags on this page.
Who maintains jeneric/STIG-MCP?
+
jeneric/STIG-MCP is maintained by jeneric. The last recorded GitHub activity is dated 2026-10-01, with 1 open issues.
Are there alternatives to STIG-MCP?
+
Yes. On ClaudeWave you can browse similar mcp servers at /categories/mcp, sorted by popularity or recent activity.
Deploy STIG-MCP to your cloud
Ship this repo to production in minutes. Each platform spins up its own environment with editable env vars.
Maintain this repo? Add a badge to your README
Drop the badge into your GitHub README to show it's tracked on ClaudeWave. Each badge links back to this page and reflects the live Trust Score.
[](https://claudewave.com/repo/jeneric-stig-mcp)<a href="https://claudewave.com/repo/jeneric-stig-mcp"><img src="https://claudewave.com/api/badge/jeneric-stig-mcp" alt="Featured on ClaudeWave: jeneric/STIG-MCP" width="320" height="64" /></a>More MCP Servers
Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
An open-source AI agent that brings the power of Gemini directly into your terminal.
Real-time global intelligence dashboard. AI-powered news aggregation, geopolitical monitoring, and infrastructure tracking in a unified situational awareness interface
🕷️ An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl! Don't be shy, join here: https://discord.gg/EMgGbDceNQ and follow here for daily tips and tricks: https://x.com/Scrapling_dev
The fastest path to AI-powered full stack observability, even for lean teams.