Transient
Back to Home
Transient IntelligenceHosted MCP

Transient Intelligence — MCP Integration

Canonical production flow: connect to the hosted MCP endpoint with OAuth, then bind your personal TI API key once with ti_set_api_key.

1) Connection values

Copy these values exactly into your MCP client or connector setup.

MCP endpoint

https://mcp.transientintelligence.com/mcp

OAuth client ID

ti-mcp-public-20260306

2) Copy/paste connector config

Use this structure for any client that accepts MCP endpoint + OAuth config.

Hosted MCP config (generic JSON)
{
"mcpServers": {
"transient-intelligence": {
"url": "https://mcp.transientintelligence.com/mcp",
"authorization": {
"type": "oauth",
"client_id": "ti-mcp-public-20260306"
}
}
}
}
Connector fields (when UI asks for values)
MCP URL: https://mcp.transientintelligence.com/mcp
Auth Type: OAuth
OAuth Client ID: ti-mcp-public-20260306
OAuth Client Secret: Do not enter one for your setup

3) First-run commands (required)

After OAuth succeeds, run these in order in your MCP client:

Required bootstrap commands
ti_set_api_key({ api_key: "ti_..." })
ti_api_key_status()
ti_workflow_policy()
ti_run_workflow({ question: "...", text: "..." })

Important: you should not be asked for an OAuth client secret. You only need OAuth sign-in and your own TI API key.

4) Quick verification checklist

Use this to confirm a new integration is correctly configured in under 2 minutes.

  1. 1.Connector saves successfully with https://mcp.transientintelligence.com/mcp.
  2. 2.OAuth sign-in opens and completes without client-secret prompt.
  3. 3.ti_api_key_status() reports key is not set on first run.
  4. 4.ti_set_api_key succeeds with your own TI key.
  5. 5.ti_run_workflow returns citations for a test prompt.

5) Core tool sequence

Recommended call order
ti_workflow_policy()
ti_run_workflow({ question, text | input_url })
// or manual: ti_upload -> ti_run_status -> ti_ask

6) Session management (important)

Reusing a session_id across unrelated reviews can mix old and new context.

  • Reuse a session only for follow-up questions on the same document set.
  • Start a new session for a new document pack, review objective, or customer/case.
  • Store session_id per job/workflow, not as a single global default.
  • Never reuse sessions across tenant or project boundaries.

Need client-specific screens?

See AI Clients for ChatGPT/Claude setup notes, and Developers dashboard for live values.