Workforce MCP

Specialist context for your agent.

One MCP. When you need UI work, load UI context. Pipeline work → DE. Reliability → SRE. Your agent gets full production-grade skills, discovery questions, stack defaults, quality bars, and handoffs for that specialty — without installing separate skill packs per project.

Security

Stdio-only MCP. No install scripts. No network / shell / eval in our published file. Packs are embedded at build time (no runtime fs). The official MCP SDK is a normal dependency so schema tooling stays upstream — see SECURITY.md.

Related MCP server: ContextAtlas

Mental model

You: call MCP prompt "workforce/DE" for the order pipeline
  → agent loads Data Engineer specialist context
  → agent works AS a data engineer on that task

Same for UI, FE, BE, ML, AI, OPS, SRE, MON, SEC, QA, ARCH, MGR, …

Orchestration flow: discuss → delegate → one specialty at a time.
Pods: workforce/WEB (UI+FE+BE), workforce/DP (DE+DS), workforce/AIP (AI+ML+DS+DE) — roster presets, not mega-skills. Specialty AI ≠ pod AIP.

Install (Cursor)

Use npx (works for everyone — no local node_modules required):

{
  "mcpServers": {
    "workforce": {
      "command": "npx",
      "args": ["-y", "@saaalil/workforce-mcp"]
    }
  }
}

Do not set command to node with ./node_modules/@saaalil/workforce-mcp/... unless you have already run npm i @saaalil/workforce-mcp in that same workspace. That path is why installs fail with MODULE_NOT_FOUND.

Or run: npx -y @saaalil/workforce-mcp.

Local clone:

{
  "mcpServers": {
    "workforce": {
      "command": "node",
      "args": ["/absolute/path/to/Workforce-MCP/dist/index.js"]
    }
  }
}

Short flags

Also accepts: workforce/DE, workforce-UI, devops, o11y, data engineer, etc.

Pods (roster presets)

Pods run member POVs + a delegation table, then you execute one workforce/FLAG. See docs/ADR-0001-pods.md.

Tools

Default mode=ask: investigate the repo, then reply with Goal / Blocking questions (0–3 with defaults) / Assumptions / Plan and stop until approved (unless the change is trivially small).

Example prompts to your agent

• workforce/UI — design the marketing landing page

• workforce/WEB — plan a web feature across UI → FE/BE

• workforce/DP — plan a data mart with DS + DE

• workforce/AIP — plan RAG/agent work (not the same as workforce/AI alone)

• workforce/DE — design the orders gold mart

• workforce_as with role SRE — define SLOs for checkout

• workforce/discuss — scrum the “express checkout” idea across specialties

• workforce/discuss with format postmortem_theater — full cast; each specialty owns one corrective action

• workforce/delegate — break “express checkout” into specialty-owned slices

• workforce/MGR — stay in manager mode and sequence the work

• Plain language also works: “Load DE context and design the orders gold mart”

• Switch context OPS → SRE after the golden path is ready

How Workforce differs from skill directories (e.g. skills.sh)

skills.sh is a marketplace of discrete skills you install one-by-one (npx skills add …) — procedural snippets for agents (frontend-design, tdd, azure-*, etc.). That model wins at breadth and remixing many tiny capabilities.

Workforce is a different bet:

Be better, not broader: stay opinionated on orchestration and bars; don’t become another skill dump. Complement skills.sh (use a skill for a niche procedure) while Workforce owns who speaks, who owns the slice, and when to switch.

Develop

npm install
npm run validate-packs
npm run build
npm run smoke

License

MIT

Support

Having install or connection issues? Reach out:

• X: @HiremathSalil

• Email: salilhiremath2712@gmail.com

• GitHub issues: https://github.com/Saaalil/Workforce-MCP/issues

Docs site: https://workforce-website-psi.vercel.app/support