OneChat MCP server

Control OneChat from Claude Code, Claude Desktop, Cursor or any other MCP client: everything the admin UI does — agents, knowledge, rules, skills, MCP connections, workflows, platforms and the support console — is available as MCP tools, scoped to your own tenant.

Endpoint

https://stg.onechat.tabsystems.tech/mcp

Transport: Streamable HTTP. Every request needs an Authorization: Bearer och_… API key header.

1. Create an API key

Log in to the admin console and open your profile.
In the API keys card press New key, name it after the client (e.g. "Claude Desktop") and optionally set an expiry.
Copy the och_… value immediately — it is shown only once. Revoke it from the same card any time.

The key inherits your operator role and tenant: every tool call acts as you, inside your tenant only.

2. Connect your client

Claude Code (CLI)

claude mcp add --transport http onechat https://stg.onechat.tabsystems.tech/mcp \
  --header "Authorization: Bearer och_…your-key…"

Claude Desktop / Cursor / any .mcp.json client

{
  "mcpServers": {
    "onechat": {
      "type": "http",
      "url": "https://stg.onechat.tabsystems.tech/mcp",
      "headers": {
        "Authorization": "Bearer och_…your-key…"
      }
    }
  }
}

Replace och_…your-key… with the key from step 1.

3. First steps

Ask your client to run the setup_agent prompt to create an agent end-to-end, or import_knowledge to bulk-load documents.
The resource docs://onechat/usage-guide carries the full reference an AI agent can read on its own.
Safe to try first: list_agents, list_platforms, send_platform_chat (a sandbox — nothing reaches real customers).

Tools

Area	Tools
Agents	`list_agents · get_agent · create_agent · update_agent · delete_agent`
Knowledge	`list_knowledge_documents · get_knowledge_document · create_knowledge_document · update_knowledge_document · reindex_knowledge · delete_knowledge_document`
Rules & skills	`list_rules · get_rule · create_rule · delete_rule · list_skills · get_skill · create_skill · delete_skill`
MCP connections	`list_mcp_servers · add_mcp_server · update_mcp_server · delete_mcp_server · test_mcp_server · sync_mcp_server`
Agent tooling	`list_builtin_tools · list_agent_workflow_tools · set_agent_workflow_tools · list_agent_connections`
Workflows	`list_workflows · get_workflow · create_workflow · update_workflow · delete_workflow · run_workflow · list_workflow_executions · assist_workflow`
Platforms	`list_platforms · get_platform · create_platform · update_platform · delete_platform · test_platform · send_platform_chat · regenerate_verify_token · regenerate_notify_webhook`
Support console	`list_conversations · get_conversation · send_message · pause_conversation · resume_conversation · reset_conversation · delete_conversation · list_escalations · list_notifications`
Tenant	`get_tenant_settings · update_tenant_settings · list_operators`

Credentials (platform tokens, MCP headers) are write-only: tools accept them but never return them. send_message delivers to a real customer — send_platform_chat is the sandbox. Conversation reads are limit-capped.