Best Tools for Teams Without a Platform Team in 2026

You have a small team. You want an assistant that can check a payment, open a ticket, or read a CRM record. You do not have a platform group whose job is the plumbing under that assistant. You should not hire three people before the first useful call.

This article names the tools that let you skip building that plumbing, what each one still leaves on your desk, and what it cost in October 2026. Swytchcode is one of the tools. It is not the only one, and for a plain catalog it is often not the first one.

The work a platform team usually owns

After a connector works in a demo, four jobs remain. A platform team does these. A small team either buys them or discovers them during an incident.

1. The login. Stripe's secret, the CRM token, and the token for each customer have to live somewhere that is not the chat and not the git repo. When a token expires in the middle of a run, someone refreshes it.
2. The allow list. "We connected Stripe" is not "the assistant may refund." The list of actions that may run is a decision. A sentence in the prompt is not that decision. A later message can talk an assistant out of a sentence.
3. The retry. A read can be tried again. A create, if the second request looks new, becomes a second charge, a second email, or a second ticket. The rule has to be "same key, same write," or "do not retry, look it up."
4. The log. When something goes wrong next month, you need the record of what ran and what was blocked. Some free plans delete that history in a week.

Building this yourself means you control every timeout. It also means your team is on call for it. Not building it is the right trade for a lot of small teams. The trade should be named so you know what you still decide.

Publishing your own APIs is a different job. That shortlist is 10 Best Enterprise API Management Platforms. Governing the call the agent makes is 10 Best API Governance Tools for Enterprise Teams.

How to evaluate a tool when you have no platform team

How the tools compare

1. Zapier MCP

Best for: The person on the team who already lives in Zapier and wants Claude or ChatGPT to use those same apps.

This is the shortest path if "no platform team" also means "no terminal." You connect an MCP URL to the Zapier account you have. Each successful tool call uses 2 tasks. Failed calls are free. Free is 100 tasks a month. A looping agent spends that in an afternoon.

Zapier MCP is built around one person's connected accounts. It is awkward when every customer of your product must connect their own Gmail. Internal and legacy APIs are not the catalog's strength. Streamable HTTP is the transport. There is no stdio option in the product as described in October 2026.

You still decide which Zapier actions are turned on, and you still do not get a documented "same refund, not a second refund" rule from the task meter.

2. Composio

Best for: Many known SaaS apps, many end users, and a free tier that does not run out during a pilot.

Under 100,000 tool calls a month, Composio's free tier is the best in this category. October 2026 also listed 50,000 trigger events, unlimited connected accounts, and 3 seats, with your own OAuth app. Managed apps can be lower. Pro is $29 a month.

A small team should start here when the assistant's job is the apps everyone already knows, and the failure you fear is "we cannot connect them," not "a retry paid someone twice." If the failure you fear is the duplicate payment, put that call somewhere that will not mint a new write. The two products can sit next to each other. You do not have to choose only one.

Write idempotency was not described in the public docs in October 2026.

3. Arcade

Best for: The assistant must act as a specific teammate or customer, and a missing permission should start a grant instead of failing silently.

Free is 2,000 tool calls and 2,000 auth events. Team is $25 a month plus $0.10 per auth event and $0.01 per tool call. At a million calls that usage is about $10,025. A small team will not hit a million calls in month one. They will hit the auth-event meter if every teammate reconnects during setup.

Arcade can deploy to their cloud, your VPC, or an air-gapped setup on Enterprise. That last option is a security conversation, not a weekend install. Idempotency on writes was not described in the public docs in October 2026.

4. Nango

Best for: You are shipping a product, and customers connect their own accounts, and someone on the team can own a TypeScript function.

Nango is more product-engineering than "our assistant should read the CRM." Free stops at 10 connections. Pay-as-you-go is $50 a month, then per connection, per compute hour, and per gigabyte. The free self-hosted edition covers auth and proxy. You operate the database, upgrades, and token storage. That is platform work. If the reason you are here is that you do not have a platform team, self-hosting Nango is the wrong row.

The cloud product is reasonable when one developer can review the function and you want the OAuth screen in your app. It is a poor fit for "we need a ticket lookup by Friday and nobody wants a new service to run."

5. Pipedream

Best for: Prototyping a workflow. Not for telling a customer the bot is live.

Free has no production environment. Startup is $99 a month billed annually or $150 billed monthly. A small team that ships on the free plan will find out in production that the plan was never a production plan. Read that line on their pricing page before you put a date on a launch.

6. n8n

Best for: A workflow the whole team can open and read, on infrastructure you already run.

n8n is a workflow builder. You draw the steps. That is the thing a lot of small teams are trying not to maintain. It is the right tool when someone enjoys that canvas and data residency rules out a hosted catalog. It is the wrong tool when the requirement is "no workflow to build."

MCP exposure has limits that matter: multi-step forms and human-in-the-loop executions were not supported over MCP in the product description we use. You still host upgrades.

7. Swytchcode

Best for: One or two expensive calls, including an internal API, when you want the login, the allow list, and the retry rule without operating a token database.

Swytchcode is a production runtime. The model decides the call. Swytchcode runs it.

On every call it does eight checks:

1. Allow list. Only an action you added with swy add method can run.
2. Schema check. Inputs are checked before the request is sent.
3. Policy. policies.json can allow, deny, or hold, and can read the arguments. You still write the rule that says a refund over a limit needs a person. The hold itself is on Business and Enterprise. Free and Pro can allow or deny.
4. Credentials. swy auth connect stores the secret outside the project. The assistant does not see it. There is no project env file.
5. Execution. The request goes from your machine to the provider. No proxy hop.
6. Retry. Backoff on rate limits. POST and PATCH retry only when an idempotency key is already set.
7. Response check. A success status whose body says the action failed is not success.
8. Audit. Local log. Free keeps 7 days, Pro 30, Business 90.

You install a CLI. That is a terminal. If nobody on the team will open a terminal, start with Zapier MCP instead, and come back when a call is expensive enough to be worth the install.

Free is 100 live commands a month, 10 connected end users, no card. Pro is $29 a month or $290 a year for 10,000 live commands. Business is $149 a month or $1,490 a year for 1,000,000, plus approval and 5 seats. If nobody approves within 48 hours, the call never runs. An internal API can be added from an OpenAPI spec or a Postman collection, which is the case catalogs are weakest at. The pricing page has the current numbers.

What you do not give up: the decision of which actions are allowed. Leave delete, pay, and invite off until you can name them.

Which tool to pick

Where the free plans stop, in more detail, is Best Free Tools for Building AI Agent Integrations.

How we checked these facts

Prices, task math, and free-tier limits are from each vendor's public pricing in October 2026. The "not described in public docs" line is the October 2026 docs check for write idempotency on Arcade, Composio, and Nango. Swytchcode limits are from swytchcode.com/pricing the same month.

Follow these steps to try the call layer

You do not need a platform team for these steps. Skip them if you picked Zapier and you are not installing a CLI.

1. Follow this step to install the CLI

npm install -g swytchcode

On Windows, in PowerShell:

irm https://cli.swytchcode.com/install.ps1 | iex

On macOS or Linux:

curl -fsSL https://cli.swytchcode.com/install.sh | sh

Then swy --version. The rest of the install is the CLI quickstart.

2. Follow this step to sign in

swy login

The first two live calls can run before an account. Details are in the authentication docs.

3. Follow this step to start a test project

swy init --editor=claude --mode=sandbox

Use cursor or codex in place of claude if that is your editor. Project setup is in Set up a project.

Then pick one service you already know, download it with swy get, and search with swy discover scoped with --project. Use the name it prints. Turn on one read. Do not turn on every action the catalog offers.

FAQ

Can a small team ship an AI agent without a platform team?

Yes, if a product stores the login, you still choose which actions are allowed, and you know what a retry does. Skipping the allow list is how a demo assistant learns to refund.

Should we build the auth and retry layer ourselves?

Build it when the call is unusual and you have someone to own it. Install or buy it when the work is the same login, retry, and log every team rebuilds. You still write the rule for the expensive action.

What does the Swytchcode free plan include?

In October 2026: 100 live commands a month, 10 connected end users, 7 days of logs, and allow or deny. No card, no team seats, no approval workflow. It is for evaluating the pipeline.

Pipedream is free. Can we use it in production?

No. The free plan is development only. It does not include a production environment.

We only need Stripe and a helpdesk. Do we need all of this?

Use a catalog for the helpdesk if that is the fast path. Use a runtime that reuses an idempotency key for the Stripe write. They can run side by side.

Read next

• Explore this article on free-tier limits: Best Free Tools for Building AI Agent Integrations.
• Explore this article if the agent will call many systems: Best Options for Agents Calling 50+ APIs.
• The difference between an API you publish and an API the agent calls is in 10 Best Enterprise API Management Platforms.

Start with one service you already understand

Set up https://www.swytchcode.com/skills.md and build a one-action test agent for a team with no platform group. Use sandbox mode. Turn on one read with swy discover. Do not enable refunds, deletes, or emails. Explain which login you stored and confirm it is not in the chat.

If you are not sure whether you are about to build a platform or buy a weekend of setup, ask. Discord is the quick place. The contact form is there if the system is internal.