ODOCK.AI
Plugins

Evaluate A Plugin

Practical checklist for deciding whether to enable a plugin.

Evaluate A Plugin

Evaluate a plugin before enabling it because plugins can participate in runtime enforcement, transformation, external integrations, and evidence collection.

Use this checklist for built-in, marketplace, private/internal, and custom-developed plugins.

1. Define The Requirement

Write the requirement in one sentence:

Use this plugin to <allow, block, transform, enrich, export, notify, or analyze> <traffic or evidence> for <scope>.

Examples:

RequirementGood plugin fit?
Export high-risk request decisions to the security team's SIEM.Yes
Add tenant metadata to selected model calls.Yes
Send an email when an enterprise approval is required.Yes
Detect special words for product analytics and recommendations.Yes
Limit requests per minute.No, use guardrail policies.
Block prompt injection.Usually no, use SafetySec.
Stop traffic after monthly spend is exceeded.No, use budgets.

2. Confirm It Belongs In Plugins

Use standard sections when possible:

3. Review Lifecycle Moments

QuestionWhy it matters
Does it run before upstream work?It may affect what leaves Odock.
Does it run after upstream response?It may affect what the caller receives.
Does it run after response/evidence?It should usually be non-blocking audit, analytics, webhook, or email work.
Can it block?Operators need a user-visible error reason and rollout plan.
Can it transform?Operators need to know which fields can change.

The plugin should run only where it has the right context. If a marketplace listing does not explain lifecycle placement clearly, do not enable it without review.

4. Review Capabilities

Confirm whether the plugin requires:

  • request inspection
  • response inspection
  • request mutation
  • response mutation
  • external network calls
  • webhook delivery
  • email delivery
  • secret or credential access
  • background work
  • model, MCP, API key, team, tenant, or organisation scope

Least privilege means granting only what the plugin needs for the approved requirement.

5. Review Data Handling

Ask:

  • Does any prompt, response, metadata, header, token count, cost, or user identifier leave Odock?
  • Which external system receives it?
  • Is the payload minimized?
  • Is sensitive data redacted or classified before export?
  • Are data residency, retention, and vendor requirements acceptable?
  • Does the plugin support an observe-only or dry-run mode?

For MCP traffic, also review MCP Security, because tool calls can carry operational data or trigger side effects.

6. Review Observability

A plugin should produce enough evidence to operate it:

EvidenceUse
Request idCorrelate plugin behavior with gateway usage.
DecisionSee whether the plugin allowed, blocked, transformed, emitted, or failed.
ScopeKnow which organisation, team, API key, model, MCP server, or tenant was affected.
External delivery idReconcile webhook, email, SIEM, DLP, or analytics exports.
Error and retry stateDistinguish plugin failure from intended enforcement.
TimingUnderstand latency added by response-path plugins.

For current custom plugin delivery, use Request A Custom Plugin. For the planned self-service marketplace model, see Marketplace.

7. Plan Rollout

Start narrow:

  • one test API key
  • one team
  • one tenant
  • one model or MCP server
  • observe-only before block, when available
  • non-production before production for high-impact plugins

Enablement Checklist

Before enabling, confirm:

  • purpose and owner are clear
  • plugin type is understood
  • lifecycle moments are appropriate
  • required capabilities are acceptable
  • data handling is approved
  • configuration values are ready
  • compatibility is confirmed
  • user-visible behavior is documented
  • observability is sufficient
  • rollout scope and rollback criteria are defined
  • Odock assistance is requested if private implementation details are needed

On this page