EXTERNAL AI CONTROL · MCP + API

Connect your AI assistant to Swipies.

An assistant with authorized HTTPS or network-enabled code can register its own separate Free workspace and perform your task directly. It can also reuse your existing authorized account. No manual MCP installation is required for the direct HTTP path.

Для ИИ: прочитайте полную инструкцию подключения. Для управления нужны MCP или API-инструменты в вашем ИИ-клиенте.

AI yordamchilari uchun: to‘liq ulash ko‘rsatmalarini o‘qing. Klient MCP yoki API vositalarini qo‘llashi kerak.

Full agent instructions Discovery JSON Live server manifest

Connect ChatGPT with OAuth

For an eligible ChatGPT account with custom remote MCP support, connect once and then describe your task in chat. Your client and administrator control availability and tool approvals.

Check the live manifest: mcp.oauth.enabled must be true.
Add a remote MCP connection in your AI client with OAuth authentication and this exact URL: https://api.swipies.app/api/v1/agent-access/mcp.
On the Swipies consent page, choose a separate Free workspace or sign in and connect your current owner account. No copied API key is needed.

В ChatGPT нужно один раз добавить MCP-подключение с OAuth и подтвердить доступ. На странице Swipies можно создать отдельную Free-область или подключить свой текущий аккаунт. Затем агент работает через инструменты подключённой платформы.

If the flag is false or absent, OAuth has not been enabled on the running server. Visiting a webpage cannot install tools in your chat. An assistant that already has authorized HTTPS/code tools can use the direct flow below.

Start directly: your agent registers over HTTPS

Give your assistant a task and the Swipies instructions. With authorized HTTP tools or network-enabled code, it can do the following itself:

Read the live manifest.
Privately generate and persist a random identity for a new workspace, or reuse authorized access to your existing account.
Call POST https://api.swipies.app/api/v1/agent-access/bootstrap to obtain a separate Free account and API key.
Send authenticated JSON-RPC to https://api.swipies.app/api/v1/agent-access/mcp, discover the operations and execute the task.

Агент регистрируется сам. Если у него есть разрешённые HTTPS-инструменты или код с сетевым доступом, ручная регистрация и настройка MCP для этого пути не нужны. Он сохраняет свою личность и повторно использует тот же аккаунт. Для существующего Plus/Pro нужен доступ именно к этому аккаунту.

Agent o‘zi ro‘yxatdan o‘tadi. Ruxsat etilgan HTTPS vositalari yoki tarmoqqa kira oladigan kodi bo‘lsa, bu yo‘l uchun qo‘lda ro‘yxatdan o‘tish va MCP sozlash shart emas. Agent shaxsiy kalitini saqlab, keyingi safar o‘sha akkauntga qaytadi.

The agent must keep credentials private and follow its host permissions. A session that can only read webpages cannot install tools by visiting this page. Free workspaces are separate and include Swipies attribution.

Read the complete executable request sequence

Optional: configure a host MCP connection

The generator below is for clients that need a configured MCP server. It is not a prerequisite when your assistant already has direct HTTPS tools.

Connect your AI client

Set up once, then describe your tasks in chat.

Setup language

1. Save the client

Download Python client

Requires Python 3.10+ and your installed AI client. Save the download as swipies_agent_client.py, removing any final .txt extension. Review the source before connecting.

2. Choose your client

Your terminal Python command

Absolute path to the saved .py file

Use the path on the machine where your AI client runs. The website cannot locate or verify local files.

Command for your selected terminal

Enter an absolute path ending in .py to generate your configuration.

Run the command in your selected terminal. Check the connection with codex mcp list.

Official client instructions

Existing account or Plus/Pro? Configure its SWIPIES_API_KEY securely in the MCP process environment before first use; follow the account guide. Without a key, first use creates a separate Free workspace. Never enter your key on this page.

Account and credential guide

3. Ask your AI to check the connection

Read https://swipies.app/llms.txt. Check the configured Swipies connection and list available dataset operations. Before creating a new workspace, confirm whether an existing account connection is configured.

Your AI client may ask you to enable or approve the tools. This page generates configuration; it does not install software or check whether you are connected.

1. Choose your workspace

Existing customer: use the existing account API key through your client's secret settings (SWIPIES_API_KEY for the stdio client). This keeps your datasets, agents and subscription.

New workspace: the direct HTTPS bootstrap or stdio client can create an isolated Free service account automatically. It saves a random identity locally and reconnects to the same account on later runs. No browser registration or copied API key is required for this flow.

A newly registered account cannot access your other accounts or inherit Pro. Keep the local identity file private and backed up. This flow has no email recovery or account claiming.

2. Connect an MCP-capable client

Requires Python 3.10+; no third-party Python packages. Download the client to your machine and inspect it using your host's normal policy.

Download swipies_agent_client.py — if your browser saves a .txt extension, rename the file to swipies_agent_client.py.

Use the equivalent of this configuration in a host with stdio MCP support. Replace the path with the real absolute local path; use python3 if needed. The host controls its configuration format and tool permissions.

{
  "mcpServers": {
    "swipies": {
      "command": "python",
      "args": [
        "/absolute/path/swipies_agent_client.py"
      ]
    }
  }
}

Codex CLI:

codex mcp add swipies -- python /absolute/path/swipies_agent_client.py

Optional client flags: --url https://api.swipies.app, --state-file /private/path/identity.credential, --name "My AI workspace". Give independent customers separate state files or user profiles. Windows credentials use DPAPI; POSIX credentials are stored owner-only.

Direct HTTP instead of stdio

With an existing key, point a host supporting bearer-authenticated Streamable HTTP MCP to https://api.swipies.app/api/v1/agent-access/mcp. Configure Authorization: Bearer <your API key> securely and send User-Agent: SwipiesAgent/1.0. The endpoint uses POST, advertises protocol 2025-03-26, and does not offer server-initiated SSE sessions. GET/DELETE return 405.

An assistant with HTTP tools can also use the documented bootstrap and REST catalog. Read the full instructions before creating an identity; persist the generated secret before the first request.

3. Give your assistant the task

For example: “Create a knowledge dataset from the documents I provide, build a support chatbot and verify its answers against the sources.”

Discover with swipies_list_operations.
Read the selected API docs with swipies_describe_operation.
Execute using swipies_execute and the discovered operation ID.
Wait for parsing, test the result and retain the created resource IDs.

Available operations come from your running server and account. Do not guess operation IDs or retry an uncertain write before checking its result.

What is available?

Agents and chatbots	Create, read, update, run and delete resources; inspect sessions and replies.
Knowledge and RAG	Create datasets, upload authorized documents, track parsing and retrieve source-grounded answers.
Configuration	Use available model, channel and connector operations with the required credentials and permissions.
Current exclusions	Administration, payments, token management, human login, OAuth callbacks and Ads are not in the generic MCP catalog.

Plans and boundaries

Free outputs include Swipies attribution. Only active persisted Plus and Pro plans can omit it. Account authorization, quotas and host permissions always apply. Private databases, model providers and OAuth services require valid authorized access; reading a page cannot supply those credentials or install a connector in every AI product.

MCP limits: 24 MiB request, 16 MiB combined uploads (20 files), 8 MiB result, 90-second operation timeout. Use smaller pages or direct REST for larger files and long-lived streams. Parsing is asynchronous.

Read the integration as an AI

Start with /llms.txt. Follow /llms-full.txt for setup and execution, discovery JSON for capabilities and URLs, and OpenAPI for the four gateway endpoints.

Verified on 2026-10-27: stdio/HTTP MCP, resource CRUD, uploads, parsing, retrieval, RAG references, ordinary/streaming OpenAI-compatible replies, tenant isolation and Free/Pro attribution. Actual external CRM delivery and paid advertising delivery remain outside this live verification.