Connect your agent to Robbclan

Work on projects, learning, and team workflows through your AI assistant. Connect over Model Context Protocol (MCP), discover the tools available to your account, and review proposed work before approving the next step.

Set up your connection

1. Sign in

Open Preferences → AI & MCP. You need a Robbclan account with access to the applications you want to use. Your assistant acts with your account permissions.

2. Create a personal MCP token

Under Personal MCP Access, give the token a recognizable label and select the appropriate portfolio and agent profile. Copy the token when issued: it is shown once. Store it in your client's protected credential settings, outside chat prompts, screenshots, shared documents, and source control.

3. Configure your client

• Server URL: https://robbclan.com/mcp
• Connection: hosted HTTP MCP
• Authentication: personal bearer token
• Header: Authorization: Bearer YOUR_PERSONAL_MCP_TOKEN

Use the client-specific configuration shown in your signed-in AI & MCP preferences. The application provides examples for Claude Code, the Anthropic API MCP connector, and OpenClaw through a local proxy. These configuration examples are not a claim of tested compatibility across every client or version. An OAuth-only remote client cannot connect using only this URL and a bearer token.

Existing connections using https://wayfinder.robbclan.com/mcp can retain that address. Use the canonical Robbclan endpoint for new connections. External provider sign-in and consent remain separate setup steps.

4. Make your first call

Call robbclan_list_workflows on the connected Robbclan server and summarize the workflows it reports. Do not create, modify, approve, publish, or send anything.

A successful response reports the workflow map from the connected server. Use MCP tools/list to inspect current schemas before selecting an action. A listed tool does not guarantee that your account can execute it. Tenant administrators can additionally call robbclan_connection_health; it does not test external clients or model providers.

Explore the tools

• Ahoy: Turn selected Slack conversations into reviewed project suggestions and reply drafts. Examples: ahoy_import_thread, ahoy_get_review, ahoy_edit_suggestion
• Passage: Create a project, generate a proposed plan, and review tasks and dependencies before approval. Examples: odyssey_create_project, odyssey_generate_plan, odyssey_project_history
• Wayfinder: Develop learning paths, course outlines, and course content through staged review. Examples: create_learning_path, get_course_outlines, get_course_pending_reviews
• Workflow: Define reusable selected-thread team workflows and inspect their saved definitions and runs. Examples: workflow_create, workflow_get, workflow_runs
• Safe Harbor: Investigate observed agent activity through collected traces, audit events, and usage summaries. Examples: safe_harbor_list_traces, safe_harbor_audit_log, safe_harbor_cost_summary
• Tenant AI: Inspect configured model routing and test supported platform routes and private connections. Examples: tenant_ai_get_routing, tenant_ai_test_route, tenant_ai_private_connection_status

Safe Harbor administration and Tenant AI require authorized tenant-admin access. Passage retains the odyssey_* names for compatibility.

Try a project-planning session

Illustrative prompt sequence, not a transcript or guaranteed result:

1. Ask: “Create a restricted draft Passage project for a team onboarding checklist. Generate a proposed plan and show the saved proposal and revision. Leave it unapproved. Do not execute tasks or send messages.”
2. Review the saved proposal. Request revisions as needed; revising does not approve it.
3. After reviewing the current saved version, explicitly approve that revision if you want to proceed. Your assistant checks that it is still current, invokes the approval tool, and reads back the recorded result. A revision conflict requires another review.

Plan approval does not itself execute tasks or authorize messages. Generation, approval, publication, sending, payments, and access changes are separate actions. Approval tools record the authenticated caller's decision; your client must obtain consent and enforce tool-approval policies.

Know what is covered

Safe Harbor covers instrumented activity and compatible submitted events, not every action of an external agent. Cost estimates are not provider invoices. Tenant AI routing applies to supported platform tasks; it does not automatically make every external assistant action private.

Troubleshooting and disconnecting

• Authentication fails: check the direct endpoint and bearer-token configuration. Replace a lost or revoked token in AI & MCP preferences.
• An action is unavailable: refresh tool discovery and check product access, role permissions, and provider connections.
• Generation times out: read the current project or generation status before retrying; the server request may still be running.
• A revision conflict occurs: fetch and review the latest saved proposal before approving again.
• Disconnect: remove the server from your client and revoke its personal token in Robbclan. Provider consent and separately issued connector credentials have their own revocation controls.

Access and pricing

Availability depends on your organization and account permissions. Contact sales@robbclan.com for current commercial terms.

Set up your connectionExplore agent documentation

Read Markdown