You have written this OAuth flow before
Token storage, refresh handling, pagination, rate-limit backoff, error mapping: the same week of work, once per service, forever. VectorHub does it once so you can go back to the product.
Per-service integration work, itemised
None of it is hard. All of it is mandatory, and none of it is your product.
- Register an app and get through the consent screen
- Store the token somewhere that is not a dotfile
- Handle refresh before it expires mid-request
- Map their pagination scheme onto yours
- Back off correctly on their rate limits
- Translate their error codes into something actionable
- Notice, six months later, that a field was deprecated
# what it looks like through the gateway
vectorhub connect github --auth oauth
vectorhub keys create ci-agent --connectors github
# that's it: no token store, no refresh loop,
# no pagination shim, no error mapping
vectorhub logs tail --key ci-agent
12:04:31 github.pr.list 142ms ok
12:04:33 github.pr.review.create 198ms okBuilt to be scripted
Everything in the dashboard is available from the CLI and the API, because integration plumbing belongs in version control.
CLI first
Connect apps, mint keys, convert APIs and tail logs from the terminal. The whole workflow scripts into CI.
Environments
Separate connections and keys per environment, so a dev agent cannot reach production Salesforce.
Typed SDKs
TypeScript and Python clients for provisioning, key management and log access, generated from the same OpenAPI spec we publish.
Local bridge
vectorhub mcp stdio proxies a local stdio client to the gateway, so development sees exactly what production sees.
Hand-rolled vs. the gateway
Both work on day one. They diverge sharply on day ninety.
Developer questions
Is there a CLI, or is everything in a dashboard?
Both. The CLI covers connecting apps, creating and revoking keys, converting APIs and tailing call logs, so the whole workflow scripts cleanly into CI.
What do the SDKs actually do?
They cover provisioning rather than tool calls: creating keys, managing connections, reading logs. The tool calls themselves go through MCP from whatever client you are using.
Can I run this locally during development?
Yes. `vectorhub mcp stdio` bridges a local stdio transport to the gateway, so a local client sees the same tools with the same scoping as production.
How do I keep dev and prod separate?
Environments are first-class. Each has its own connections and its own keys, so a dev agent pointed at a sandbox Salesforce cannot reach the production one.
Go back to building your product
Connect the apps, mint the key, and never write another refresh-token loop.
No credit card required · Free tier available · Self-host on request