Private beta

Tremi MCP

Connect a developer MCP client to discover published workflows, inspect normalized run evidence, and create contained simulations.

Streamable HTTP endpointhttps://www.tremi.ai/api/mcp

Private beta access is workspace-specific.A Tremi platform administrator must enable MCP before the workspace can create MCP-scoped keys or connect to this endpoint.

1
Enable MCP
Ask a Tremi platform administrator to enable MCP for the workspace.
2
Publish a workflow
Publish the workflow whose contract and simulation tools the client will use.
Open workflows
3
Create a dedicated key
Grant only workflows:read, workflows:simulate, and runs:read.
Open API keys
4
Connect and verify
Add the endpoint and bearer key to Codex or MCP Inspector, then confirm all five tools appear.
Configure a client

Successful connection5 tools discovered

list_published_workflows
get_workflow_contract
simulate_workflow
list_runs
get_run

Access and authentication

Use a dedicated Tremi bearer key and grant only the scopes the client needs.

Scope	Allows
`workflows:read`	List published workflows and read their safe contracts.
`workflows:simulate`	Create deterministic, internally contained simulations.
`runs:read`	List runs and read normalized run evidence.

workflows:run remains a backward-compatible simulation superset, but live runs are not available through MCP.

Codex

Store the Tremi key in TREMI_API_KEY, then add the server to your Codex configuration.

Codex config.toml

[mcp_servers.tremi]
url = "https://www.tremi.ai/api/mcp"
bearer_token_env_var = "TREMI_API_KEY"

MCP Inspector

Start MCP Inspector with npx @modelcontextprotocol/inspector.
Select Streamable HTTP and enter https://www.tremi.ai/api/mcp.
Add Authorization: Bearer $TREMI_API_KEY, connect, and confirm the five tools listed above.

Toollist_published_workflows

List published workflows

List published Tremi workflows available in the authenticated workspace.

Scope: workflows:read
Read only: true
External delivery: None

MCP tool

list_published_workflows

list_published_workflows input

{
  "limit": 20
}

list_published_workflows result

{
  "workflows": [
    {
      "name": "Evaluation follow-up",
      "slug": "evaluation-follow-up",
      "description": "Follow up after a customer evaluation.",
      "version": 4,
      "publishedAt": "2026-08-27T14:12:00.000Z"
    }
  ],
  "nextCursor": null
}

Input

Field	Type	Required	Description
`cursor`	`string`	No	Opaque cursor returned by the previous page.
`limit`	`integer`	No	Page size from 1 through 50. Defaults to 20.

Result

Field	Type	Required	Description
`workflows`	`array`	Yes	Published workflow summaries in stable publication order.
`workflows[].name`	`string`	Yes	Workspace-facing workflow name.
`workflows[].slug`

Toolget_workflow_contract

Get workflow contract

Get the safe public contract for one published Tremi workflow.

Scope: workflows:read
Read only: true
External delivery: None

MCP tool

get_workflow_contract

get_workflow_contract input

{
  "workflowSlug": "evaluation-follow-up"
}

get_workflow_contract result

{
  "name": "Evaluation follow-up",
  "slug": "evaluation-follow-up",
  "description": "Follow up after a customer evaluation.",
  "version": 4,
  "schemaVersion": 2,
  "channels": [
    "voice"
  ],
  "stopConditions": [
    "voice.goal_completed",
    "recipient.opt_out"
  ],
  "requiredDestinations": [
    "phone"
  ],
  "variables": {
    "first_name": {
      "type": "string",
      "description": "Customer first name.",
      "nullable": false,
      "sensitive": false,
      "example": "User"
    }
  },
  "publishedAt": "2026-08-27T14:12:00.000Z"
}

Input

Field	Type	Required	Description
`workflowSlug`	`string`	Yes	Slug of a published workflow in the authenticated workspace.

Result

Field	Type	Required	Description
`name`	`string`	Yes	Workspace-facing workflow name.
`slug`	`string`	Yes	Stable published workflow slug.
`description`	`string`	Yes	Safe published workflow description.
`version`	`integer`

Toolsimulate_workflow

Simulate workflow

Create a contained deterministic simulation that stores normal Tremi run evidence.

Scope: workflows:simulate
Read only: false
External delivery: None

MCP tool

simulate_workflow

simulate_workflow input

{
  "workflowSlug": "evaluation-follow-up",
  "idempotencyKey": "evaluation-2048",
  "variables": {
    "first_name": "User"
  }
}

simulate_workflow result

{
  "id": "run_01K4MCP2048",
  "state": "queued",
  "workflowVersion": 4,
  "createdAt": "2026-08-27T14:12:00.000Z",
  "replayed": false,
  "resultWebhookDelivery": "suppressed"
}

Input

Field	Type	Required	Description
`workflowSlug`	`string`	Yes	Slug of a published workflow in the authenticated workspace.
`idempotencyKey`	`string`	Yes	Caller key of at most 200 characters for safe retries.
`variables`	`object`	No	Scalar workflow variables. Defaults to an empty object.

Result

Field	Type	Required	Description
`id`	`string`	Yes	Public identifier of the simulated run.
`state`	`run state`

Toollist_runs

List runs

List normalized Tremi run summaries in the authenticated workspace.

Scope: runs:read
Read only: true
External delivery: None

MCP tool

list_runs

list_runs input

{
  "workflowSlug": "evaluation-follow-up",
  "mode": "simulation",
  "limit": 20
}

list_runs result

{
  "runs": [
    {
      "id": "run_01K4MCP2048",
      "workflow": {
        "name": "Evaluation follow-up",
        "slug": "evaluation-follow-up",
        "version": 4
      },
      "state": "completed",
      "outcome": "goal_achieved",
      "mode": "simulation",
      "resultWebhookPolicy": "suppress",
      "createdAt": "2026-08-27T14:12:00.000Z",
      "updatedAt": "2026-08-27T14:14:06.000Z"
    }
  ],
  "nextCursor": null
}

Input

Field	Type	Required	Description
`workflowSlug`	`string`	No	Return runs for one workflow slug.
`state`	`run state`	No	Return runs in one lifecycle state.
`mode`	`live \| simulation`	No	Return runs created in one execution mode.
`cursor`	`string`	No	Opaque cursor returned by the previous page.
`limit`	`integer`	No	Page size from 1 through 50. Defaults to 20.

Result

Field	Type	Required	Description

Toolget_run

Get run

Get normalized evidence for one Tremi run without recipient or provider-sensitive data.

Scope: runs:read
Read only: true
External delivery: None

MCP tool

get_run

get_run input

{
  "runId": "run_01K4MCP2048"
}

get_run result

{
  "id": "run_01K4MCP2048",
  "workflow": {
    "name": "Evaluation follow-up",
    "slug": "evaluation-follow-up",
    "version": 4
  },
  "state": "completed",
  "outcome": "goal_achieved",
  "mode": "simulation",
  "resultWebhookPolicy": "suppress",
  "attempts": [
    {
      "stepIndex": 0,
      "channel": "voice",
      "state": "delivered",
      "errorCode": null,
      "startedAt": "2026-08-27T14:12:02.000Z",
      "completedAt": "2026-08-27T14:13:26.000Z"
    }
  ],
  "events": [
    {
      "name": "run.created",
      "occurredAt": "2026-08-27T14:12:00.000Z"
    },
    {
      "name": "run.completed",
      "occurredAt": "2026-08-27T14:14:06.000Z"
    }
  ],
  "createdAt": "2026-08-27T14:12:00.000Z",
  "updatedAt": "2026-08-27T14:14:06.000Z"
}

Input

Field	Type	Required	Description
`runId`	`string`	Yes	Public Tremi run identifier.

Result

Field	Type	Required	Description
`id`	`string`	Yes	Public run identifier.
`workflow`	`object`	Yes	Workflow name, slug, and immutable version. Draft simulations use version 0 and include draftRevision.
`state`	`run state`	Yes	Current run lifecycle state.
`outcome`	`outcome \| null`	Yes

Simulation isolation

An MCP simulation creates ordinary Tremi run evidence without causing external delivery.

Tremi creates a normal run in simulation mode and stores deterministic attempts, run events, state transitions, and the terminal outcome.
Tremi generates reserved synthetic recipient values and fixed {source: "tremi-mcp"} metadata server-side.
No provider adapter is invoked and no recipient is contacted.
No result-webhook event, attempt, delivery job, or outbound webhook request is created.
The result reports resultWebhookDelivery: "suppressed".

External delivery is off by default.MCP and REST simulations do not send result webhooks. In the console, explicitly select an endpoint to send test results.

Optional MCP test-webhook delivery is deferred to a later release. V1 has no input or separate tool that can enable it.

Pagination and idempotency

List tools use opaque cursors backed by stable timestamp and public-identifier ordering.

List size defaults to 20 and accepts values from 1 through 50.
Pass the returned nextCursor unchanged to retrieve the next page.
simulate_workflow requires an idempotency key of at most 200 characters.
Tremi namespaces that key to the authenticated API key. Replaying the same key and input returns the original run.
Reusing the key with different input returns idempotency_conflict.

Workflow contracts

Published workflow metadata gives agents a useful contract without exposing prompts, message bodies, provider resources, or run input values.

Schema v2 requires a workflow description and a descriptor for every referenced template variable. Each descriptor declares type, description, nullable, sensitive, and an optional example. Publishing rejects missing or extra descriptors and run creation enforces declared types and nullability.

Published schema-v1 workflows remain readable and runnable. Tremi represents each inferred v1 template variable as any_scalar, nullable, with a generic description. Saving and publishing a new revision upgrades the authoring contract to v2.

Errors and limits

Tool failures are returned as safe MCP error results without raw exceptions or request data.

Code	Meaning
`authentication_required`	The bearer key is missing or invalid.
`forbidden`	The API key lacks the tool scope.
`mcp_not_enabled`	The workspace private-beta entitlement is disabled.
`validation_failed`	The tool input, cursor, workflow variables, or contract is invalid.
`not_found`	The workflow or run is not available in the authenticated workspace.
`idempotency_conflict`	The simulation key was reused with different input or is still processing.
`rate_limited`	The key exceeded 120 protocol requests or 10 simulations per minute.
`internal_error`	The request could not be completed safely.

Tool annotations

Tool annotations tell compatible clients how each operation behaves.

Tool	readOnlyHint	destructiveHint	openWorldHint
`list_published_workflows`	true	false	false
`get_workflow_contract`	true	false	false
`simulate_workflow`	false	false	false
`list_runs`	true	false	false
`get_run`	true	false	false

simulate_workflow is not read-only because it creates a run. It is non-destructive and has openWorldHint: false because it cannot contact providers, recipients, or result-webhook destinations.

ChatGPT integration

The private beta uses Tremi API keys for developer MCP clients.

Public ChatGPT web integration requires the later OAuth phase and is not supported by this bearer-key beta. See the OpenAI authentication guidance for the authorization model required for that release.

`runs`	`array`	Yes	Normalized run summaries in stable creation order.
`runs[].id`	`string`	Yes	Public run identifier.
`runs[].workflow`	`object`	Yes	Workflow name, slug, and immutable version. Draft simulations use version 0 and include draftRevision.
`runs[].state`	`run state`	Yes	Current run lifecycle state.
`runs[].outcome`	`outcome \| null`	Yes	Terminal outcome when one is available.
`runs[].mode`	`live \| simulation`	Yes	Execution mode captured by the run.
`runs[].resultWebhookPolicy`	`deliver \| suppress`	Yes	Immutable result-webhook policy.
`nextCursor`	`string \| null`	Yes	Cursor for the next page, or null when the page is final.

Tremi MCP

Enable MCP

Publish a workflow

Create a dedicated key

Connect and verify

Access and authentication

Codex

MCP Inspector

Input

Result

Input

Result

Input

Result

Input

Result

Input

Result

Simulation isolation

Pagination and idempotency

Workflow contracts

Errors and limits

Tool annotations

ChatGPT integration