Claude Financial Services: what the public plugins actually provide
Deploying Financial Services: plugin or managed-agent wrapper?
Choose a host, preserve configuration and keep preview orchestration outside production approval paths.
What you will learn
- Match deployment to the operator
- Inventory configuration and dependencies
- Design rollback before a real connection
Before you start
- Command-line and JSON familiarity
- A disposable workspace with synthetic financial records
A proposed learning project turns validated requests into reviewable records without giving documents control of routing.
Key takeaways
- Preview delegation needs an operational owner.
- File-based content and Python harnesses have different requirements.
- Rollback includes in-flight work.
Match deployment to the operator
Cowork installation targets a person working through the host interface. Managed-agent templates target an application with its own workflow engine. Both reuse agent material, but the second route adds credentials, session routing and operational responsibility.
A team that cannot yet name the owner of failed sessions should begin with a supervised plugin trial. The README identifies callable_agents as a preview capability; a reference wrapper is insufficient evidence for unattended production reliability.
Inventory configuration and dependencies
Record the repository revision, host version, enabled plugin directories, connector identities and data-provider permissions. Store secrets through the chosen host secret mechanism, outside reviewed manuscripts, shell history and example configuration committed to Git.
The repository advertises file-based plugin content without a build step, while the managed examples include Python and SDK calls. Treat those as separate deployment surfaces. Do not invent a Docker service for the plugin files or assume the orchestration sample installs its dependencies.
Design rollback before a real connection
Preserve the previous plugin or wrapper revision and a sanitized configuration export. Pilot the new version with an isolated dataset; compare missing-field handling and reviewer decisions before routing new work to it.
Rollback must also address in-flight work. Pause new submissions, retain draft evidence and decide which tasks need replay with an idempotency key. Reinstalling old instructions alone does not undo a provider-side action or duplicate session.
Decision guide
| Criterion | Option A | Option B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
Implementation steps
- 1
Select supervised plugin or application-owned wrapper.
- 2
Record versions, connector scopes and secret locations.
- 3
Test restoration with synthetic work before rollout.
Copy-ready example
revision: 574ed3624aebd0418c7e96cd101262f30210ab26
mode: supervised-plugin-trial
data: synthetic
provider_secrets: external-secret-store
rollback: previous-reviewed-plugin-copyFrequently asked questions
Does the repository deploy an entire financial platform?
No. It provides workflow artifacts and reference integration material around a host and data services.
Is the managed API verified here?
No live API deployment was performed; preview behavior must be checked in your environment.
Sources
- Financial Services / README.mdSource checked 2026-09-23
- Financial Services / scripts/orchestrate.pySource checked 2026-09-23