Connect an agent
Give a coding agent its own key: pick its projects, decide who starts its split tests, and pause or disconnect it any time.
Connecting an agent takes two minutes. Open Agents, press Connect an agent, and pick which one: Claude Code, Cursor, Codex, or any other.
To connect a coding agent, you make it a key in Needlr and paste one line into the agent. Owners and editors can do this (who's who).
Think of a guest pass
A guest pass at an office opens only the rooms you choose. It stops working on a set date, and you can take it back any time.
An agent's key is a guest pass to Needlr. You choose its projects, how much it may do on its own, and when it stops.
Before you start
- An authenticator app on your phone. Creating an agent needs a code from it, and Needlr helps you set one up if you don't have it (why).
- At least one project with your website (how).
How to connect one
- 1Open Agents in the menu on the left, and press Connect an agent.
- 2Which agent: pick Claude Code, Cursor, Codex, GitHub Copilot, Devin, CI pipeline, or Other agent. This only changes the setup lines you get at the end.
- 3Name (so you know which one it is): for example “Claude Code, website repo”. Leave it empty and Needlr uses the agent's name.
- 4Projects it can work on: tick one or more. The agent sees only these.
- 5Its split tests start…: pick one of the three answers below. There's no default; you have to choose. Only the account Owner can pick the two that start on their own; an Editor connects agents that always ask.
- 6Press Connect, and type the code from your authenticator app if Needlr asks for one.
- 7Copy the key right away. Needlr shows it only once, with the exact lines for your agent: a command for Claude Code, a file for Cursor, or a ready request for a CI pipeline.
Who starts its split tests
- Always ask me. Its split tests wait until someone starts them in Needlr. They show under Waiting for you, with Review and start.
- On its own, when the code came through a pull request. Someone can review the code before it's live. Without a pull request, it asks.
- Always on its own. Its split tests start as soon as it asks. You're responsible for what it ships.
This choice is only about split tests. Measuring a release that's already live, or how many people use a new feature, never changes your site. Those start on their own, as soon as Needlr is counting their KPI.
Every option on an agent's card
- Pause turns the agent's keys off until you press Resume. Impact checks that are already measuring keep measuring.
- New key makes a second key. The old one keeps working until you press Revoke next to it, so you can swap keys without a gap. Revoke shows when an agent has more than one key.
- Disconnect turns the agent off for good. Its keys stop working at once. Its impact checks and their history stay.
- Its split tests start lets you change the choice above at any time.
- Keys lists the start of each key and when it expires. A key lasts 90 days; make a new one before then.
- Verdicts sent to: press Add a webhook and give an address that starts with https. Needlr then posts each verdict there, signed with a secret it shows once. After repeated failed deliveries it turns off. Without one, the agent picks up its verdicts next time it asks.
What needs a code from your phone
Giving an agent more power needs a code from your authenticator app: connecting it, New key, a webhook, Resume, and letting its split tests start on their own.
Taking power away never needs one: Pause, Revoke, Disconnect, removing the webhook, or switching back to Always ask me.
For developers: the address and the API
- Agents connect over MCP (Streamable HTTP) at https://api.needlr.io/mcp, with the header Authorization: Bearer followed by the key.
- The same actions work as plain HTTP under https://api.needlr.io/v1/agent/, which is what the CI pipeline option uses.
- Each key can make up to 120 calls a minute. Every refusal comes with next steps the agent can follow.
- Add one line to your agent's instructions file so it uses Needlr at the right moments. The setup card gives you that line.
Common mistakes
- Not copying the key. It's shown once. If you lost it, press New key, then Revoke the old one.
- Giving an agent every project. Pick only the ones it actually works on.
- Forgetting the instructions line. Without it the agent has the tools but doesn't know when to ask.
What the video says
Connecting an agent takes two minutes. Open Agents, press Connect an agent, and pick which one: Claude Code, Cursor, Codex, or any other.
Choose the projects it can work on. It sees only those projects, their KPIs and their impact checks, and nothing else in your account.
Then your call: do its split tests start on their own? Always ask me, only when the code came through a pull request, or always. Your choice, your responsibility.
Needlr shows the key once, with the exact line to paste into your agent. Copy it right away. Every key expires after 90 days.
Pause it, issue a new key, or disconnect it at any time. Giving an agent more power takes a code from your phone. Taking power away never does.
From then on, your agent asks before it builds and after it ships, and reads its own verdicts. You see every impact check it asks for, and you keep the final say.