Installation by client
The @shiftleftpt/sbd-toe-mcp server is distributed exclusively via npm (plus an alternative GitHub Release bundle) and runs as a stdio process — compatible with any standard MCP client.
Prior requirements​
| Requirement | Detail |
|---|---|
| Node.js | ≥ 20.9.0 |
| Access to the npm registry | Public — no token needed |
| Disk space | A few tens of MB (snapshot of the Manual, KG and ontology) |
| Network | Only on first start (npx does the fetch) or on upgrades |
For environments without npm access, download the bundle from the GitHub Release, verify the .sha256 and reference the extracted dist/index.js via command: "node" (see the section GitHub Release Bundle).
Claude Code (CLI)​
The simplest form — one command, zero file editing:
claude mcp add sbd-toe -- npx -y @shiftleftpt/sbd-toe-mcp
To verify:
claude mcp list
To remove:
claude mcp remove sbd-toe
The configuration is stored in ~/.claude.json (global scope) or in .mcp.json (project scope, with --scope project).
Claude Desktop (macOS / Windows)​
Edit the file claude_desktop_config.json:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"sbd-toe": {
"command": "npx",
"args": ["-y", "@shiftleftpt/sbd-toe-mcp"]
}
}
}
Restart Claude Desktop. The sbd-toe.* tools appear in the conversation's tools panel.
Cursor​
Edit ~/.cursor/mcp.json (global) or .cursor/mcp.json in the repository (per project):
{
"mcpServers": {
"sbd-toe": {
"command": "npx",
"args": ["-y", "@shiftleftpt/sbd-toe-mcp"]
}
}
}
Under Settings → Features → MCP, confirm that sbd-toe appears as running (green badge).
VS Code + GitHub Copilot​
Add .vscode/mcp.json to the repository (auto-detected by Copilot Chat in Agent mode):
{
"servers": {
"sbdToe": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@shiftleftpt/sbd-toe-mcp"]
}
}
}
Alternatively, global user configuration via Settings UI → Extensions → GitHub Copilot → MCP.
Windsurf (Codeium)​
Edit ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"sbd-toe": {
"command": "npx",
"args": ["-y", "@shiftleftpt/sbd-toe-mcp"]
}
}
}
Zed​
Edit ~/.config/zed/settings.json, section context_servers:
{
"context_servers": {
"sbd-toe": {
"command": {
"path": "npx",
"args": ["-y", "@shiftleftpt/sbd-toe-mcp"]
}
}
}
}
Other MCP clients​
Any client that supports the stdio transport of MCP uses the same pattern: command: "npx" + args: ["-y", "@shiftleftpt/sbd-toe-mcp"]. The client's documentation gives the location of the configuration file.
GitHub Release Bundle​
For environments without access to npm (air-gapped, self-hosted) or for pinning to a specific version:
- Download the bundle from github.com/SbD-ToE/sbd-toe-mcp/releases —
sbd-toe-mcp-v<versão>-bundle.tar.gz(or.zip) — and its.sha256, and verify the checksum before extracting. - Extract to a known path —
/opt/sbd-toe-mcp/for example. - Reference
dist/index.jsinstead ofnpx:
{
"mcpServers": {
"sbd-toe": {
"command": "node",
"args": ["/opt/sbd-toe-mcp/dist/index.js"]
}
}
}
Determining the project's risk level​
Without the correct risk level, MCP returns a set of controls that is needlessly broad or dangerously narrow. The level is computed with the Ch. 01 method (exposure, data and impact axes, E+D+I). The indicators below are illustrative hints, neither necessary nor sufficient, and the regulatory context is declared separately: a regime does not imply L3.
| Indicator (illustrative) | Typically |
|---|---|
| Internal application with low impact (even with personal data, if E+D+I ≤ 4) | L1 |
| Public APIs or processing of user data | L2 |
| Public exposure with sensitive data and high impact | L3 |
The level decision follows the method of Ch. 01 — no tool takes it on the project's behalf. The tool map_sbd_toe_applicability requires riskLevel and accepts technologies, hasPersonalData, isPublicFacing and projectRole: it does not decide the level, but it shows the effect of each level on the project, which helps compare the alternatives when in doubt.
Final verification​
Whatever the client, validate with:
list_sbd_toe_chapters()
It should return 15 chapters (00-fundamentos to 14-governanca-contratacao). If so — it is operational.
Next​
Configure a skill / agent file so that the AI client consults the Manual automatically instead of requiring the user to ask explicitly.