Python SDK

The Python SDK wraps every v3 endpoint as a typed function, generated from the same OpenAPI spec that powers the API Reference. Install it, call configure() once, and call the operation functions. The same package also ships an optional MCP server.

pip install assistableai
# or: uv add assistableai

Requires Python 3.10+.

Configure

import assistableai
from assistableai import api

client = assistableai.configure(
    api_key="ask_live_…",          # or env ASSISTABLE_API_KEY
    subaccount_id="<subaccount-id>",  # optional default X-Subaccount-Id
    # base_url defaults to https://api.assistable.ai
)

configure() sets Authorization: Bearer <api_key> and (when provided) X-Subaccount-Id on every request, and remembers the client process-globally. Arguments fall back to the environment (ASSISTABLE_API_KEY, ASSISTABLE_SUBACCOUNT_ID, ASSISTABLE_BASE_URL). Prefer an explicit instance? Use AssistableAI(...) directly — it takes the same arguments and exposes the same .raw client.

from assistableai import AssistableAI

client = AssistableAI(api_key="ask_live_…", subaccount_id="<subaccount-id>")

Make a request

Operations live under api.<topic>.<operation>. Pass client=client.raw and the operation’s parameters. Every operation returns the full v3 envelope — { data, error, request_id } — as a typed model:

result = api.assistants.list_assistants.sync(client=client.raw, limit=2)

if result.error:
    print(f"Request {result.request_id} failed:", result.error)
else:
    for assistant in result.data:
        print(assistant.id, assistant.name)

Path parameters are positional; query params and bodies are keyword arguments:

assistant = api.assistants.get_assistant.sync("<assistant-id>", client=client.raw)

contact = api.contacts.create_contact.sync(
    client=client.raw,
    body={"first_name": "Ada", "email": "[email protected]"},
)

Each operation also exposes:

sync_detailed(...) — returns the HTTP status code and headers alongside the parsed body.
asyncio(...) / asyncio_detailed(...) — await-able variants for async code.

result = await api.assistants.list_assistants.asyncio(client=client.raw, limit=2)

The import path from assistableai._client.api.assistants import list_assistants also works if you prefer importing operations directly.

Per-call subaccount override

Pass x_subaccount_id= to target a different subaccount for a single call without changing the global default:

api.assistants.list_assistants.sync(client=client.raw, limit=10, x_subaccount_id="<other-subaccount>")

MCP server

The package bundles an MCP server that exposes every v3 operation as a tool. Install the optional extra:

pip install "assistableai[mcp]"

Run it directly to sanity-check your env:

ASSISTABLE_API_KEY=ask_live_… assistableai-mcp
# or: python -m assistableai.mcp

Add it to an MCP client (Claude Desktop, Cursor, Claude Code, …). Using uvx means no separate install step:

{
  "mcpServers": {
    "assistableai": {
      "command": "uvx",
      "args": ["--from", "assistableai[mcp]", "assistableai-mcp"],
      "env": {
        "ASSISTABLE_API_KEY": "ask_live_…",
        "ASSISTABLE_SUBACCOUNT_ID": "<subaccount-id>"
      }
    }
  }
}

Variable	Required	Purpose
`ASSISTABLE_API_KEY`	yes	Bearer key (`ask_live_…`)
`ASSISTABLE_SUBACCOUNT_ID`	no	Default `X-Subaccount-Id` for every tool call
`ASSISTABLE_BASE_URL`	no	Override (default `https://api.assistable.ai`)

Auth is applied on the underlying HTTP client, so it reaches every tool call. The server registers one tool per operation (e.g. listAssistants, createContact, createCall).

The typed surface (operation functions + models) is generated from spec/openapi.json and kept in lockstep with the API. See the API Reference tab for the full list of operations and their shapes.

Getting Started

SDKs & Tools

Guides

Configure

Make a request

Per-call subaccount override

MCP server

​Configure

​Make a request

​Per-call subaccount override

​MCP server

Configure

Make a request

Per-call subaccount override

MCP server