Tool Profiles

Tool profiles control which tools appear in MCP discovery for a connection. They reduce schema overhead and help a client select the right tool, but they do not replace OAuth scopes, workspace roles, confirmation, or server-side authorization.

Questions This Page Answers

Which profile should I use for a client or operator workflow?
What happens when no profile, or an unknown profile, is requested?
How are private connections different from provider directory publication?

How to Use Profiles

Append ?profile=NAME to the hosted MCP URL:

https://mcp.useorgx.com/mcp?profile=commander

Or include the profiled URL in a client configuration:

{
  "mcpServers": {
    "orgx": {
      "command": "npx",
      "args": ["mcp-remote", "https://mcp.useorgx.com/mcp?profile=executor"]
    }
  }
}

The default is v2, the compact public surface represented by the checked-in OrgX MCP server manifest. Missing and unknown profile names fail closed to v2. The full profile is available only when explicitly requested for an admin or debugging compatibility session.

Client and Directory Status

Surface	Endpoint	Current meaning
General remote MCP clients	`https://mcp.useorgx.com/mcp`	Uses the default `v2` surface
ChatGPT Developer Mode	`https://mcp.useorgx.com/mcp?profile=chatgpt`	Available for private testing; not proof of OpenAI directory publication
Claude custom connector	`https://mcp.useorgx.com/mcp`	Available as a user-added custom connector
Claude Code direct MCP	`https://mcp.useorgx.com/mcp`	Available through `claude mcp add` and OAuth login
Anthropic review profile	`https://mcp.useorgx.com/mcp?profile=claude-directory`	Available in production for authenticated review testing; not proof of directory publication
Provider directories	Provider-managed listing	Submission, approval, and publication are separate from endpoint availability

The dedicated claude-directory profile is live for authenticated review testing. Endpoint availability does not mean the connector was submitted, approved, or published in Anthropic’s directory. Continue to use the bare /mcp endpoint for ordinary Claude custom and direct connections.

Current Profiles

The current profile definitions come from the OrgX MCP server’s src/toolProfiles.ts. Reconnect a client after changing a profile URL so it refreshes tools/list.

`v2` (Default)

The public server-manifest surface. It contains the canonical OrgX workflows, widget affordances, and supported client-integration tools. Use the bare https://mcp.useorgx.com/mcp URL to select it implicitly.

`chatgpt` (23 tools)

The focused reviewer surface for private ChatGPT Developer Mode testing:

orgx_bootstrap, orgx_search, orgx_inspect, orgx_recommend, orgx_write,
orgx_attach, orgx_act, manage_lifecycle, orgx_plan, orgx_spawn, orgx_decide,
orgx_submit_receipt, approve_decision, reject_decision, get_agent_status,
get_initiative_pulse, scaffold_initiative, handoff_task, approve_agent_work,
review_artifact, get_morning_brief, get_operator_chronicle,
check_execution_readiness

Within this 23-tool surface, the informational tools have these review annotations:

readOnlyHint: true (5): orgx_inspect, review_artifact, get_morning_brief, get_operator_chronicle, check_execution_readiness
readOnlyHint: false because successful calls consume metered MCP allowance (4): orgx_search, orgx_recommend, get_agent_status, get_initiative_pulse

`claude-directory` (7 tools)

The focused, non-destructive, closed-world Anthropic review surface:

orgx_search, orgx_inspect, orgx_recommend, get_agent_status,
get_initiative_pulse, get_morning_brief, get_operator_chronicle

Its review annotations are:

readOnlyHint: true (3): orgx_inspect, get_morning_brief, get_operator_chronicle
readOnlyHint: false because successful calls consume metered MCP allowance (4): orgx_search, orgx_recommend, get_agent_status, get_initiative_pulse

`memory` (7 tools)

Shared organizational memory, context, decisions, attachments, and receipts:

orgx_bootstrap, orgx_search, orgx_inspect, orgx_decide, orgx_recommend,
orgx_attach, orgx_submit_receipt

`commander` (21 tools)

Human operator workflows for inspection, action, planning, delegation, attention and question handoffs, execution readiness, and receipts:

orgx_bootstrap, orgx_search, orgx_inspect, orgx_recommend, orgx_write,
orgx_attach, orgx_act, orgx_plan, orgx_spawn, orgx_decide,
orgx_request_attention, orgx_poll_attention, orgx_ack_attention,
orgx_request_question, orgx_poll_question, orgx_emit_execution_graph,
orgx_submit_receipt, check_execution_readiness, scaffold_initiative,
consolidate_pr, get_operator_chronicle

`planner` (12 tools)

Initiative planning, hierarchy scaffolding, decisions, and human handoffs:

orgx_bootstrap, orgx_plan, orgx_write, orgx_act, orgx_search, orgx_inspect,
orgx_decide, orgx_request_attention, orgx_poll_attention,
orgx_request_question, orgx_poll_question, scaffold_initiative

`executor` (18 tools)

Agent execution, telemetry, handoffs, changes, delegation, receipts, and PR consolidation:

orgx_bootstrap, orgx_emit_activity, orgx_request_attention,
orgx_poll_attention, orgx_ack_attention, orgx_request_question,
orgx_poll_question, orgx_emit_execution_graph, orgx_search, orgx_inspect,
orgx_write, orgx_attach, orgx_act, orgx_plan, orgx_spawn,
check_execution_readiness, orgx_submit_receipt, consolidate_pr

`observer` (9 tools)

A narrow monitoring and reporting discovery surface:

orgx_bootstrap, orgx_search, orgx_inspect, orgx_recommend, orgx_plan,
orgx_decide, orgx_poll_attention, orgx_poll_question,
get_operator_chronicle

A narrow discovery profile is not necessarily uniformly read-only. readOnlyHint reflects observable side effects, including metered-usage writes. OAuth scopes, workspace roles, confirmation policy, and server authorization remain authoritative for every tool call.

`full` (Explicit admin/debug only)

All tools registered by the server, including compatibility and internal surfaces. It is never selected by omission or an unknown profile name. Use it only when an admin or debugging workflow genuinely requires the complete catalog.

Choosing a Profile

Use case	Recommended profile
General remote MCP connection	`v2` through the bare `/mcp` URL
Private ChatGPT Developer Mode test	`chatgpt`
Authenticated Anthropic review test	`claude-directory`
Shared company memory	`memory`
Human operator session	`commander`
Initiative planning	`planner`
Background agent execution	`executor`
Focused monitoring	`observer`
Explicit admin/debug compatibility	`full`

Start with the narrowest profile that covers the workflow. If a required tool is absent, choose a better-matched focused profile before reaching for full.

Getting Started

Connect Your Tools

Agent Ops

Platform

API Reference

Resources

Questions This Page Answers

How to Use Profiles

Client and Directory Status

Current Profiles

`v2` (Default)

`chatgpt` (23 tools)

`claude-directory` (7 tools)

`memory` (7 tools)

`commander` (21 tools)

`planner` (12 tools)

`executor` (18 tools)

`observer` (9 tools)

`full` (Explicit admin/debug only)

Choosing a Profile

Next Steps

Agent Quickstart

Full Tool Reference

​Questions This Page Answers

​How to Use Profiles

​Client and Directory Status

​Current Profiles

​v2 (Default)

​chatgpt (23 tools)

​claude-directory (7 tools)

​memory (7 tools)

​commander (21 tools)

​planner (12 tools)

​executor (18 tools)

​observer (9 tools)

​full (Explicit admin/debug only)

​Choosing a Profile

​Next Steps

Agent Quickstart

Full Tool Reference

Questions This Page Answers

How to Use Profiles

Client and Directory Status

Current Profiles

`v2` (Default)

`chatgpt` (23 tools)

`claude-directory` (7 tools)

`memory` (7 tools)

`commander` (21 tools)

`planner` (12 tools)

`executor` (18 tools)

`observer` (9 tools)

`full` (Explicit admin/debug only)

Choosing a Profile

Next Steps