Skip to main content Announcing Tool Policies: AI agent permissions down to the field Read the announcement

AI Connector Builder. Any API, ready for your agents.

Turn an internal system, a niche SaaS tool or a customer's custom fields into tested actions your agents call over MCP, the AI SDKs or the Actions API.

claude · ~/brightwell Claude Code

/connector:full brightwell from docs.brightwell.example/openapi.json

stackone-agent-harness · research, build, validate, each closed by a scripted gate

  • Research 38 endpoints, 6 resources, v3 is current gate passed
  • Auth OAuth 2.0 client credentials, token request works gate passed
  • Build 24 actions in 6 partial files gate passed
  • Validate 24 of 24 match the docs, stackone validate passes gate passed

test it on our staging account, then publish

  • Test 24 of 24 actions pass on a real account 1 fixed, re-run
  • Publish brightwell v1.0.0 pushed to your project

Companies already connected through StackOne

Two ways to build, or ask us. Your agent or ours, on the same engine.

Build in Claude Code with StackOne's skills and tools, or ask StackOne's agent in the dashboard. Both run the same steps and write the same YAML. Or have StackOne's team build it for you.

  • Runs in your terminal

    Bring your own agent

    One command gives Claude Code StackOne's research, versioning and testing tools, plus the skills and slash commands that run a build. Then ask for a connector by name, or point it at the docs.

  • Runs on StackOne

    Use StackOne's agent

    Ask in the dashboard chat, to build a connector or change one you have. StackOne's agent runs the same build on our side, with nothing to install and no model bill, and asks before it touches a real account or publishes.

  • Built by StackOne's team

    Ask StackOne to build it

    Request the connector and StackOne's team builds and maintains it. Typically days, once you give us API access. Fork it afterwards if you want to change it yourself.

The AI Connector Builder is part of the Team and Enterprise plans, and OEM Enterprise. Each connector you build or change and keep active uses one custom connector slot.

How a build works. Five steps, and a script decides when each one is done.

Each step closes only when its scripted check passes, never on the model's own say. Pick a step to see the tools it called and what it wrote, and switch agents: the build is the same.

Build a connector for our shipments API. The spec is at docs.brightwell.example/openapi.json

StackOne agent Your agent Claude Code

On it. I'll start with your OpenAPI spec.

Yours to change. The connector is YAML in your project: edit it, version it and check it in CI.

Brightwell Shipments connector

  • OpenAPI 3.1
  • 6 resources
  • OAuth 2.0
  • client credentials
  • list_shipments
  • get_shipment
  • create_shipment
  • list_rates
  • list_carriers
  • list_invoices
  • validated
  • staging account
  • v1.0.0
AI Connector Builder Brightwell Shipments brightwell
stackone-agent-harness Same skills and tools in both agents.

Test and publish

Skill(test-connector)

24 actions × 1 auth type · staging account

stackone run

list_rates · 400 · missing origin_postcode

Skill(build-connector)

fix: origin_postcode is required

Skill(scramble-credentials)

no plaintext credentials in the project

stackone push

brightwell v1.0.0

test all actions · staging account

list_shipments200 · 50 rows
get_shipment200
list_rates400 · missing origin_postcode
fix: origin_postcode is required
list_rates200
create_shipment201 · removed after the test
stackone pushv1.0.0

AI Connector Builder demo. Watch a build from prompt to connector.

Change a connector that exists. Fit any connector to a customer's setup.

Most builds start from a connector that already works. Pull it, ask for the change, and push your version.

  1. Pull it

    Any StackOne connector, or one you built, comes down as YAML you can read.

  2. Ask for the change

    A customer's custom objects, a field they renamed, an action StackOne doesn't ship. Claude Code or StackOne's agent writes it the way the rest of the connector is written.

  3. Check it and push

    Validate it, run it on the customer's account, and push. It goes out as a new version for your project, and nobody else's connector changes.

  4. Customizing connectors in the docs (opens in a new tab)

Terminal

# Start from StackOne's Workday connector
stackone pull --connector workday --profile production

# Ask your agent for the change, then check and try it
cd connectors/workday
stackone validate .
stackone run --profile production \
  --connector workday.connector.s1.yaml \
  --account-id <customer account> --action-id <action id>

stackone push . --profile production

What you get. YAML you own, served to every agent.

A build leaves files in your project, not a black box. Publish them from your laptop or your CI, and every agent and app you run can call them.

Your files

  • brightwell.connector.s1.yaml
  • brightwell.shipments.s1.partial.yaml
  • brightwell.rates.s1.partial.yaml
  • brightwell.invoices.s1.partial.yaml
  • and 3 more partial files

YAML you read, review and keep in Git.

Your project

  • brightwellv1.1.0
  • brightwellv1.0.0

Each push is a new version. Validate on every pull request and publish on merge with GitHub Actions (opens in a new tab).

Your agents and apps

Every connector runs on the Agent Execution Engine, which handles sign-in, token refresh, pagination, retries and rate limits, with the same tool policies and request logs as StackOne's own connectors.

FAQ. Questions about the AI Connector Builder

Build the connector your agents are missing.

Book a demo to watch a build on an API you choose, or start free and explore StackOne's connectors.