experts-know-mcp

MCP server for Experts Know — plug your trusted human experts into your AI agent's conversations.

Bring your own lawyer, CPA, CFO, advisor, CISO, designer — anyone you already trust. Your agent invites them once, then pre-shapes work the way each expert wants to see it and hands them a structured brief whenever human review is actually needed. No more 20-page AI-generated drafts with no context.

Why

LLMs are confident about contracts, tax elections, SAFEs, privacy policies, security posture, and offer letters. They're also wrong in ways a founder can't see until it's expensive. This server gives your agent the missing escalation layer: an advisory board of the humans you already rely on, callable from inside the chat.

What your agent gets

Four moves, in order:

1. Build the board. list_my_board shows who's on it; add_expert invites your existing experts in plain language ("add my lawyer Casey at [email protected] — handles contracts and SAFEs, not litigation").
2. Decide if a human is needed. Before you sign, file, submit, send, publish, or rely on AI output, your agent calls assess_expert_review_need and gets back a domain + scope recommendation.
3. Package and route. request_expert_review turns the conversation into a structured packet (terms, assumptions, missing context, source documents, the exact question) and routes it to the right expert on your board.
4. You approve, in chat. The agent shows you a 1–2 sentence summary of what it's about to send. You say yes → approve_packet fires the actual send. You say no → reject_packet cancels. Nothing reaches an expert until you approve.

When the expert responds, the AI gets the response and rolls it into your conversation — and the approved guidance becomes reusable context for future work.

Setup

The fastest path is the agent-native bootstrap: paste this into your AI agent (Claude Code, Cursor, etc.) and let it drive setup end-to-end:

Set yourself up to use Experts Know — see https://expertsknow.com/for-agents

Or set it up manually:

1. Sign up at expertsknow.com/signup — new accounts start with $20 in free credits.
2. Copy your API key from Settings → API Keys.
3. Add the server to your MCP client config:

{
  "mcpServers": {
    "experts-know": {
      "command": "npx",
      "args": ["-y", "experts-know-mcp"],
      "env": {
        "API_URL": "https://expertsknow.com",
        "API_KEY": "your-api-key"
      }
    }
  }
}

Works with Claude Desktop, Claude Code, Cursor, Windsurf, and any MCP-compatible client. Restart your client after adding it.

Environment

| Variable | Required | Description | |-----------|----------|--------------------------------------------------------------------| | API_KEY | yes | Your workspace API key from Settings. Authenticates every call. | | API_URL | no | API base URL. Defaults to http://localhost:3000 for local dev. |

Tools

| Tool | Purpose | |------------------------------|------------------------------------------------------------------------------------------| | list_my_board | Show the experts on your board (domains, what they handle / don't, invite + active status). Use at the start of setup and before add_expert to avoid double-inviting. | | add_expert | Invite a new expert to your board. Agent extracts structured fields (name, email, domain, handles, does_not_handle) from the founder's plain-language description. | | assess_expert_review_need | Safety router. Should this work get human review? Returns recommended domain + scope. | | request_expert_review | Propose a matter; server routes against your board + builds a structured packet. Returns matter_id, packet_id, routed expert. Free under board_free — nothing reaches the expert until you approve. | | approve_packet | Fire the send-to-expert step after the in-chat approval. The agent calls this when the founder says "yes" / "send" / "approve". | | reject_packet | Terminally cancel a matter before the expert is contacted. Agent calls this when the founder says "no" / "cancel". | | check_expert_review_status | Lifecycle status, ETA, reviewer attribution, and the expert's response once complete. | | list_expert_capabilities | Domains, required inputs, supported artifacts, tiers, pricing, and SLA expectations. | | list_my_requests | Previous matters with status — avoids duplicate submissions. | | quote_expert_review | Dormant under board_free. Price/ETA/inputs for pay-per-review. | | check_credit_balance | Dormant under board_free. Credit balance + recent transactions. | | list_pricing | Dormant under board_free. Configured pricing defaults. |

Domains

Your board can include experts across:

Legal (counsel) · Tax (CPA) · Finance (CFO) · Fundraising (advisor) · Security (CISO) · HR (people ops) · GTM (advisor) · Product (advisor) · Operations (operator) · Investor (board)

Each invited expert tells you which domains they handle and which they don't, so request_expert_review routes to the right person automatically.

Spending controls

Under BILLING_MODE=board_free (default for new accounts) the agent doesn't spend anything — request_expert_review creates a matter at $0 and you approve it before any expert is contacted. Your experts are your experts; the platform just handles routing, packaging, and audit trail.

Under BILLING_MODE=pay_per_review (dormant) the legacy controls apply: max-per-review, daily/monthly caps, approval thresholds, and domain allowlists set in your dashboard. Requests that exceed a limit or hit a sensitive domain are blocked until you explicitly approve them.

Links

• Product & agent docs: https://expertsknow.com/for-agents
• Sign up / API key: https://expertsknow.com/signup
• Dashboard / audit trail: https://expertsknow.com/dashboard

License

MIT