DataHub Platform MCP · 39 Tools

DataHub Platform MCP — operate your integration platform through your Agent

Qeasy DataHub exposes assets, strategies, scripts and audit as MCP tools: a tenant user authorizes once, and the Agent operates the instance with the authorizer's identity and permissions — browse assets, inspect strategies, change configuration, manage schedules, fully auditable. Public cloud and private deployments share one implementation; keys are endpoint-self-describing, so one Agent can connect to multiple instances.

Live: 39 tools (22 core + 14 extended + 3 meta)

The authoritative count is meta.capabilities.tool_count at runtime; legacy keys without the audit:read scope see 38. This list reflects what is implemented; the Roadmap section states what is not.

meta.* — instance self-description

meta.whoami (instance/tenant identity), meta.capabilities (tool count and capability list), meta.enums. Keys carry endpoint info, enabling multi-profile agents that connect to several instances (public cloud + each private customer).

asset.platform / asset.api — API assets

List and inspect platform and API assets: available APIs, connectors and parameter shapes. Check assets before configuring strategies to avoid hallucinated fields.

strategy.* — strategy lifecycle

copy/create/update/online/offline/checkList/data.import: duplicate and create strategies, update configs, go online/offline, checklists, data import. Note: copy/create from default asset snapshots may yield empty request[].value — inspect and complete before going online.

script.create / script.delete — scripts (admin)

Create and delete transform scripts. script.test is not implemented (roadmap); locate script errors via platform logs first.

audit.search — audit (audit:read scope)

Search the audit trail of the authorizing principal. audit:read is a new scope; legacy keys do not gain it automatically.

Public cloud vs private deployment

Public cloud endpoint: https://pro-service.qliang.cloud/mcp (MCP over HTTP, Bearer sh-xxx key).

Private deployments: every customer has its own domain. The platform's “Authorize MCP” dialog generates an install guide customized to that customer's domain; wire the returned mcp.json into your Agent (self-describing keys + multi-profile, zero code changes).

Getting started (≤ 3 minutes)

1. Log in to the platform → “Authorize MCP” → pick scopes → generate an sh-xxx key (90 days by default).
2. Add the returned mcp.json to Claude Desktop / Cursor / your own agent (or npx skills add qeasy-cloud/datahub).
3. Call meta.whoami first to confirm the instance, then browse assets and strategies.

Claude Desktop / Cursor — mcp.json

{
  "mcpServers": {
    "datahub-platform": {
      "url": "https://pro-service.qliang.cloud/mcp",
      "headers": {
        "Authorization": "Bearer sh_YOUR_GRANT_KEY"
      }
    }
  }
}

Native JSON-RPC — initialize → tools/list

curl -s https://pro-service.qliang.cloud/mcp \
  -H 'content-type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -H 'Authorization: Bearer sh_YOUR_GRANT_KEY' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"my-agent","version":"1.0"}}}'

Security & trust levels

Identity is the boundary: one sh-xxx key = one tenant user's identity projection; the Agent's permissions equal the authorizer's, never more, with hard cross-tenant isolation.
Capability-level scopes: reads open by default; writes/ops/deletes authorized per capability (strategy:write, script:write, audit:read …).
Trust levels: confirm (second confirmation via confirmationToken) is implemented; guarded (human-in-the-loop IM approval) and refined full-mode policies are on the roadmap.
Every call lands in the audit trail (human → key → tool → params → result); rate limits are enforced per key.

Roadmap (not implemented — do not build against it)

strategy.delete / script.test / asset.connector.list / audit.stats — four P1 tools not yet implemented.
guarded trust level with human-in-the-loop IM approval (only the confirm flow exists today).
Known-issue fixes: empty request[].value from copy/create default snapshots; strategy.online only checks checkList ('none' is not blocking); shared test databases drop and rebuild dh_mcp_grant (use a dedicated test DB).
Private offline distribution (offline skills bundle + platform MCP Resources fallback) is scheduled with Phase 4.

Log in to the platform and issue a key to your agent

Generate an sh-xxx key and install guide from “Authorize MCP”. For knowledge retrieval, see /mcp/datahub-knowledge.

Knowledge domain guide About the website knowledge MCP