The primary startup path is Claude Code
The inspected dopemux start path launches Claude Code with workspace setup, saved project context, and configured tool coordination. Native hooks can add session context and supported workflow limits or checkpoint rules around lifecycle and tool events.
Hooks need installation and registration. They are Claude-specific, and some exceptions return an allow/fallback response. Their presence does not prove identical context delivery or approval enforcement across every client.
Tool exposure varies by client and task
The catalog contains client-specific exposure declarations and task-selected profiles. Those are projections of intended tool access, not a receipt showing an installed client can call every tool.
Some non-Claude paths are gated or limited to read-plane evidence. The current read-only facade exposes six local target-evidence tools; older broader catalog descriptions do not expand that entrypoint into a universal memory or workflow API.
Profiles help make selection explicit, while credentials, client support, identity, and live service state still need their own checks.
Project and repository connections
| Integration | Job | Current boundary |
|---|---|---|
| Leantime bridge | Projects, tickets, milestones | Operator/PM path. Sync acknowledgment does not perform bidirectional sync. |
| Official GitHub MCP | Repository, issue, PR and selected review evidence | Configured read-only; token and external image required. No posting or merging promise. |
| Serena | Symbol and code-relationship navigation | Adapter present; transport and shared-workspace claims unresolved. |
| Dopetask adapter | External task execution handoff | Pinned dependency and configured runtime required; can install dependencies. |
Keep source review separate from repository writes and external execution. The normal GitHub exposure policy excludes write tools.
Optional research and specialist tools
GPT Researcher has a configured adapter for public research. Context7 is an externally managed catalog entry for documentation lookup. Their availability depends on provider access and client exposure.
Playwright is declared for UI-audit profiles, Semgrep for security-focused work, and a repository-domain reader requires a tracked executable and safe manifest. Desktop Commander is decision-required and excluded from normal profiles. A gateway declaration does not make every connector enabled.
These are optional integration declarations. They do not establish browser publication, scan coverage, account access, or general autonomous operation.
Model routing is a configuration surface
LiteLLM provides a configured model-access endpoint with aliases, retries, and fallbacks. Alias names can differ from the actual provider configuration. Inspect the resolved route and any runtime receipt before making identity or billing claims.
Offline route reports and quota metadata help explain configuration. Current capacity, admission, provider cost, and successful requests remain separate observations.
Evaluate a connection in a narrow order
- Name the job.Choose a concrete read, write, diagnostic, or execution task.
- Inspect exposure and transport.Identify the actual client, profile, endpoint, and project scope.
- Check prerequisites.Confirm credentials, services, provider access, and intended permissions.
- Verify the result.Inspect the actual response and destination state before describing the connection as working.
This site performed source review, not those live integration steps.
Source evidence
Inspected source at the pinned reference revision. Source presence does not establish runtime health.