MCP serverNew

Workestra in Claude Desktop. In one command.

The first business OS with a public Model Context Protocol server. Claude Desktop, Claude Code, Cursor, Windsurf, VS Code, ChatGPT — any MCP client. Workspace-scoped keys, read-only mode, feature filtering.

30-second install

Run it from any terminal, or drop it into claude_desktop_config.json.

The Workestra MCP server ships as a single npm package. Pass your workspace-scoped API key. That's the entire setup.

  • Workspace-scoped keys — issue, rotate, and revoke per workspace.
  • SHA-256 hashed in the database. We never see the raw key after issue.
  • Read-only mode via --read-only for analytics use.
  • Feature filtering via --features=crm,projects.
Run anywherebash
npx -y @workestra/mcp --api-key=sk_live_your_key_here
claude_desktop_config.jsonjson
{
  "mcpServers": {
    "workestra": {
      "command": "npx",
      "args": [
        "-y",
        "@workestra/mcp",
        "--api-key=sk_live_your_key_here"
      ]
    }
  }
}
Works in

Eight clients today. Every MCP client tomorrow.

Anthropic's protocol is a 2025 open standard. Workestra is built to it directly — no custom adapters per IDE, no broken parity.

Claude Desktop

Anthropic's official desktop app — first-class MCP support.

Claude Code

Anthropic's CLI / VS Code agent. Same config file format.

Cursor

Configure once in Cursor settings, use from any project.

Windsurf

Codeium's agentic IDE. MCP-first.

VS Code

Via the MCP extension or GitHub Copilot's agent mode.

ChatGPT

MCP support for ChatGPT desktop and the OpenAI agents SDK.

Continue

Open-source coding assistant — wire Workestra into the workflow.

Any MCP client

If it speaks the Model Context Protocol, it speaks Workestra.

59 tools your AI can call

Real writes. Audited, reversible, scoped.

Every tool returns a result row that links back to the entity. Every write is in your audit log. Every key is workspace-scoped.

CRM

search_contacts · create_deal · update_deal_stage · log_call · send_sms

Support

search_tickets · triage_ticket · draft_reply · merge_tickets · breach_check

Time & Projects

log_time_entry · start_timer · create_task · move_to_cycle · approve_timesheet

Finance

create_quotation · convert_to_invoice · send_peppol_invoice · record_expense

Bookings

create_booking_link · check_availability · reschedule · cancel_booking

Knowledge

search_docs · summarise_document · cite_source · suggest_articles

Security model

Built for the threat model AI agents introduce — not retrofitted.

Workspace isolation is enforced at the database row level via Postgres RLS, not just the API. Same defenses your in-app users get.

Workspace-scoped API keys

Keys are issued per workspace. A key for Workspace A cannot read or write any data in Workspace B — enforced at the database row level, not just the application.

SHA-256 hashed at rest

We never store the raw key. Only its SHA-256 hash. Lost a key? Revoke and re-issue — no support ticket required.

Read-only mode

Pass --read-only to the MCP server (or set it on the API key) and writes are refused at the protocol layer. Safe for analytics and dashboards.

Feature filtering

--features=crm,projects exposes only those tool categories. Lock down what an AI client can touch on a per-key basis.

Use cases

Four patterns we already see in beta.

Customers use the MCP server to bring their business OS into the chat window they already keep open all day.

01

Update your CRM by chatting with Claude

"Find the deal with Acme Corp, move it to Negotiation, and add a note that they want a 12-month commit." Claude does the three writes. You approve.

02

Generate quotations from a transcript

Paste a sales-call transcript into Claude, ask for a quotation. Claude calls create_quotation with the right line items, the right contact, the right currency.

03

Log a time entry in plain English

"Log 90 minutes against the Acme onboarding project, billable, tagged 'kick-off'." Done. The entry appears in your Workestra timer history.

04

Triage tickets from your editor

From Cursor or VS Code: "Show me support tickets touching the auth module from this week, sorted by SLA breach risk." Workestra returns them — Claude triages.

Wire Workestra into your AI client

One npx away from your business OS in Claude.

Spin up a workspace, issue a key, paste it into your client's MCP config. From there, every conversation is grounded in your real data.

Free workspace. Workspace-scoped key. Revoke any time.