AI assistants (MCP)

BitAssure has an MCP server, so an AI assistant such as Claude Code, Cursor or VS Code's Copilot can use it while you work. You can ask it to:

• "Record a BitAssure journey on our staging site: sign in as the test user, add the blue T-shirt to the basket, and check the total is $24.99."
• "I changed the checkout code. Run the Checkout journey and tell me what broke."
• "What's failing this morning? Show me the screen where it failed."
• "The total is $24.99 now, on purpose. Update the check."
• "Write a BitAssure flow for our orders API from the routes in this repository, then run it."

The assistant records, writes and runs journeys the same way you would in the portal. Everything it makes is an ordinary journey you can open, edit, schedule and report on.

Connecting

The server is:

https://api.bitassure.app/mcp

There are two ways in. A chat app such as claude.ai or ChatGPT signs you in with your BitAssure account, the way any "Sign in with" button works. A developer tool such as Claude Code, Cursor or VS Code can sign in the same way, or use an API key.

claude.ai and the Claude apps

1. In Claude, open Settings → Connectors and choose Add custom connector.
2. Name it BitAssure and paste the server address above. Leave the advanced settings empty.
3. Press Connect. A BitAssure sign-in page opens: sign in, then press Yes to let Claude use BitAssure.

Claude then works as you, in your account and with your role. The connection lasts 30 days, then Claude asks you to sign in again. Custom connectors are available on Claude's paid plans; on a Team or Enterprise plan an owner may need to add it first.

ChatGPT

In ChatGPT, turn on developer mode (Settings → Apps & Connectors → Advanced settings), then create a connector with the server address above and OAuth as its authentication. Sign in and allow it as above.

With an API key

Create a key under Settings → API keys, named after the assistant, for example "Claude Code". A key acts as a tester and belongs to the account rather than to a person, which suits a shared machine or a pipeline.

Claude Code

claude mcp add --transport http bitassure https://api.bitassure.app/mcp \
  --header "Authorization: Bearer ba_your_key"

Cursor

In ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{
  "mcpServers": {
    "bitassure": {
      "url": "https://api.bitassure.app/mcp",
      "headers": { "Authorization": "Bearer ba_your_key" }
    }
  }
}

VS Code

In .vscode/mcp.json:

{
  "servers": {
    "bitassure": {
      "type": "http",
      "url": "https://api.bitassure.app/mcp",
      "headers": { "Authorization": "Bearer ba_your_key" }
    }
  }
}

Claude Code can also sign in instead: leave out --header, then type /mcp in Claude Code and choose Authenticate.

Other assistants

Any assistant that supports remote MCP servers over HTTP works: with an Authorization: Bearer header holding a key, or by signing in (MCP's OAuth sign-in) where the assistant offers it.

Keep the key out of files you commit. Most assistants can read it from an environment variable instead.

What the assistant can do

Set up an app

Tool	Does
create_project	Creates a project, one product of yours such as "Storefront"
add_platform	Adds a website or an API to a project, at its address. Phone apps need their build uploaded in the portal.

Both need an admin or owner, as in the portal, so they work when the assistant is signed in as one. A key acts as a tester and is refused.

Run and read

Tool	Does
list_journeys	Every project, platform and journey, with ids and each journey's latest result
describe_journey	A journey in words: its scenarios, steps and checks
run_journey	Starts a run of a journey or a whole platform. Spends credits.
wait_for_run	Waits up to 50 seconds for a run to finish; the assistant calls it again until it has
get_run	A run's result: what did not hold and why, where it stopped, and its link
recent_runs	A journey's latest runs, to answer "since when has this been failing?"
failing_now	Across the account: what is failing and since when, every check that did not hold, and what nothing covers
failure_picture	The screen at the moment a check failed, taken from the run's video

Look after journeys

Tool	Does
change_check	Changes what a saved check expects, keeping its history: for when the app changed on purpose
repoint_check	In a recording, points a check at the element as it is now, when it moved or was renamed
set_runs_first	Makes a journey start where others end, such as after "Sign in", so no password is typed again

describe_journey lists each scenario's and check's id, which these take.

Record a website or phone journey

Tool	Does
start_recording	Opens the recorder on a real browser or phone: a new journey, more of an existing one, or one saved scenario to edit. starts_after makes it begin where another journey ends.
look	The screen now: its address, a numbered list of what is on it, and a picture
act	One step: click, type, choose, scroll, swipe, answer a pop-up and so on, on element N
add_check	Something that should be true about element N: visible, enabled, reads, contains, attribute, on, off
remove_step, remove_check	Undo a step or a check
save_scenario	Saves what was recorded since the last save as a scenario, such as "Verify Basket"
finish	Saves or throws away anything unsaved and frees the browser or phone

The assistant picks elements by their number on the screen. BitAssure still decides how to find each element again on the next run, so a journey recorded this way is as sturdy as one you record yourself.

Write an API flow

Tool	Does
create_flow, show_flow	Start a flow on an API platform, or show one with its ids
add_call, change_call, remove_call	The calls, in order, with headers, body and {{values}}
try_call	Sends one call and shows the answer. Free.
add_check_on_answer, remove_check_on_answer	What each answer should be: status, time, body, a JSON value, a header
carry_value, remove_carried_value	Take a value from one answer for later calls, such as a token

See APIs and load testing for what each check and placeholder means.

Watching it record

When the assistant starts a recording, it gives a link to the recorder in the portal. Open it and you see the same browser or phone, with each step and check appearing as the assistant makes it. You can take over at any point: what you do and what the assistant does go into the same recording.

What it costs

The same as anything else: a website or API run is 1 credit, a phone run 1 credit a minute, and the recorder is charged by the clock while it is open (1 credit per 10 minutes on a website, 1 a minute on a phone). An assistant is slower than a person, so recording a phone through one costs more than recording it yourself. run_journey is marked as spending, so most assistants ask you before using it.

Safety

• The assistant can do what the tools offer, and no more. It can set up an app, record, write, run and read, but cannot delete journeys, change members, create keys or touch billing: none of those are tools. Signed in, it acts with your role (a viewer cannot use it, since every tool call counts as a change); with a key, as a tester.
• A sign-in given to an assistant opens the MCP server only. It cannot be used against the rest of BitAssure's API, and you can end it by removing the connector in the assistant. To end every sign-in at once, change your password under Settings → Your account and tick Sign out from other devices.
• Sign in once, then never again. Record signing in as its own journey (by hand, or with a test account), then have the assistant start other journeys after it with starts_after or set_runs_first. It records them already signed in and never needs the password.
• Use test accounts. Whatever the assistant types, it also sends to its AI provider, and what is typed into a journey is stored with it. Never give an assistant a real customer's password.
• Credentials for API platforms are set by a person under the platform in the portal. The assistant writes calls that use them but never sees them.
• Everything is reviewable. An assistant's journey is an ordinary journey: read it on its In words tab, fix it with Change, or delete it.

Limits

• Recording on an iPhone through an assistant has not been tried yet. Websites and Android have.
• "Scroll until you see…" is not offered. The assistant scrolls and then acts on what came into view.
• A phone can take a minute or two to open. start_recording then answers that it is still opening, and the assistant asks again.