OrgX MCP

Give every MCP-compatible assistant shared workspace context, accountable execution, and reviewable proof.

OrgX MCP gives every MCP-compatible assistant a shared, reviewable record of organizational work. Connect once to carry decisions, owners, delegated work, artifacts, approvals, and proof across clients and conversations.

Ask naturally. Your assistant chooses the right OrgX tool. Use this page when you want to connect, understand the work model, or find the safest next call. Use the MCP Tools Reference when you need the exact contract for every tool.

Connect

Add this remote MCP server to any compatible client:

https://mcp.useorgx.com/mcp

The hosted endpoint uses OAuth 2.1 with PKCE. Your client opens OrgX in a browser, where you sign in and choose the workspace and access you want to grant. No API key is required.

Use the streamable HTTP endpoint above for new clients. The legacy SSE endpoint is available at https://mcp.useorgx.com/sse when a client cannot use streamable HTTP. For ChatGPT Developer Mode, use https://mcp.useorgx.com/mcp?profile=chatgpt.

1. Add the endpoint

   Paste the hosted URL into your MCP client or use the client-specific setup below.

2. Approve the connection

   Complete the browser sign-in and review the workspace plus requested resource access before approving.

3. Confirm the session

   Call orgx_bootstrap. Do not start work until the returned workspace and visible tools match what you intended to authorize.

A healthy connection has a valid OAuth session, the expected workspace, a non-empty granted_scopes value, and the tools required for the current workflow in visible_tools.

Quick setup by client

ChatGPT

Enable Developer Mode in ChatGPT under Settings → Security, then add a custom connector with:

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

Complete the OrgX OAuth flow in your browser. See the ChatGPT integration guide for the ChatGPT-specific connection and widget checks.

Cursor

Add OrgX to your Cursor MCP configuration at ~/.cursor/mcp.json:

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

Restart Cursor. On first use, mcp-remote opens the browser OAuth flow.

Claude

For a Claude custom connector, add:

https://mcp.useorgx.com/mcp

For Claude Code, run:

claude mcp add --transport http --scope user orgx https://mcp.useorgx.com/mcp
claude mcp login orgx

Complete OAuth in the browser when prompted. Team and Enterprise Claude workspaces may require an Owner to add the custom connector first.

VS Code

Run MCP: Open User Configuration from the Command Palette and add:

{
  "servers": {
    "orgx": {
      "type": "http",
      "url": "https://mcp.useorgx.com/mcp"
    }
  }
}

VS Code launches the OrgX OAuth flow on first use.

OpenClaw

Install the plugin, then complete browser pairing:

openclaw plugins install @useorgx/openclaw-plugin

Open the local plugin dashboard, choose Connect OrgX, and approve the connection. The OpenClaw bridge is a local connection mode; it is not the same transport as hosted remote MCP.

OpenClaw setup guide →

For more clients, transport details, and recovery steps, see MCP Client Setup.

How OrgX works

OrgX turns a one-off prompt into a durable work episode: context → decision → execution → evidence.

Shared workspace context

Work stays inside a workspace boundary, so assistants can retrieve the right decisions, initiatives, artifacts, and active work without mixing unrelated company context.

Accountable execution

Delegated work has an owner, lifecycle, approval boundary, and evidence trail. Completion is a claim until the relevant artifact or receipt is recorded.

Object	What it holds	Why it matters
Workspace	The account and access boundary	Keeps context and permissions scoped
Initiative	A durable objective with work beneath it	Gives multi-step work a place to roll up
Task or agent run	Assigned execution and current state	Makes ownership and progress inspectable
Decision	A choice, rationale, and approval state	Preserves why the team moved forward
Artifact or receipt	The output and evidence of work	Makes “done” reviewable instead of implied

Start every session with bootstrap

Call orgx_bootstrap at the beginning of a fresh session, after reconnecting, or before a multi-step workflow:

{
  "workspace_id": "optional-workspace-id",
  "initiative_id": "optional-initiative-id"
}

The arguments are optional. The purpose is to establish context before the assistant searches, delegates, changes state, or claims progress.

Returned field	Use it for
workspace	Confirm the active workspace and account boundary
granted_scopes	Confirm the access the OAuth session actually received
visible_tools	Confirm the selected profile and scope expose the tools you need
Routing guidance	Decide whether to search, inspect, plan, delegate, or ask for approval

If the workspace is wrong or the required tool is absent, stop and reconnect or choose the correct profile and access. Do not compensate by guessing IDs or claiming work happened elsewhere.

The durable work loop

1. Start from context

   Bootstrap the session and confirm the workspace, scopes, and visible tools.

2. Find the relevant record

   Use orgx_search for a question or orgx_inspect when you already know the record ID.

3. Choose the next move

   Use orgx_recommend or orgx_plan when priority, sequencing, or acceptance criteria are unclear.

4. Act within the boundary

   Use orgx_decide, orgx_spawn, scaffold_initiative, or orgx_act according to the user’s intent and granted access.

5. Leave proof behind

   Attach the artifact or evidence with orgx_attach and record a completion receipt with orgx_submit_receipt when the work is actually done.

Tools

Your assistant chooses these tools automatically. Use this table when you need the exact routing decision, minimum input, or result to expect.

Tool	Use it for	Minimum input	Result
orgx_bootstrap	Establish session and routing context	Optional workspace or initiative ID	Workspace, scopes, visible tools, and guidance
orgx_search	Find decisions, artifacts, initiatives, tasks, or context	Query or typed filters	Matching records with enough context to choose the next call
orgx_inspect	Read one known record deeply	Record type and ID	The record plus related work, decisions, or evidence
orgx_recommend	Recover after time away or prioritize	Entity or workspace context	Ranked next actions, gaps, or blockers
orgx_decide	Record, list, approve, or reject a decision	Action plus decision fields or ID	Decision state and approval history
orgx_spawn	Assign work to a specialist agent	Agent plus task instructions	Queued work, owner, run, budget, and guardrail outcome
orgx_act	Start, block, complete, or change lifecycle state	Entity ID, action, and reason when needed	Updated state or a structured validation error
orgx_attach	Connect an artifact or evidence to work	Parent ID plus artifact or evidence	Linked artifact/evidence record
orgx_submit_receipt	Record proof of completed work	Work/run ID plus outcome and evidence	Receipt status and validation result
get_agent_status	See what agents are doing now	Optional agent or initiative filter	Running, idle, blocked, and completed work
get_initiative_pulse	Check initiative health	Initiative ID	Progress, milestones, blockers, and recent activity

The full reference is generated from the tool contracts and includes deprecated aliases, exact schemas, required scopes, profiles, and response examples. Use MCP Tools Reference when implementing a client or debugging a validation error.

Common workflows

Continue existing work

Call orgx_bootstrap, then orgx_search, orgx_inspect, and orgx_recommend. Start from the existing record instead of opening a parallel task.

Record a decision

Use orgx_decide with the choice, rationale, owner, and initiative when known. Ask for approval when the decision changes a consequential plan.

Delegate work

Use orgx_spawn only when the user wants an agent to do the work. Include the expected artifact, deadline, quality bar, and initiative or task binding when available.

Complete with proof

Do not infer completion from a queued or successful tool response. Attach the output, submit a receipt, and let the user review the resulting evidence.

Profiles and access

The endpoint profile controls discovery: it determines which tools the client sees. OAuth scopes control which resource actions the connection can perform. Workspace membership still determines which records the user can reach.

Tool profiles

Profile	Best for
v2	Default compact public surface
chatgpt	ChatGPT-compatible tools and interactive UI
memory	Search, context, decisions, and receipts
commander	Human-led planning, delegation, and approvals
planner	Initiative planning and decision handoffs
executor	Agent execution, evidence, and receipts
observer	Monitoring and reporting

Use the bare endpoint for most clients. Append ?profile=NAME only when a smaller surface helps the workflow. For example:

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

Reconnect the client after changing the profile so it refreshes tools/list.

OAuth access presets

Read

View decisions, agents, initiatives, artifacts, receipts, and memory without changing work.

Operate

Read and change work, including decisions, delegated agents, initiatives, and evidence.

Customize

Choose read or operate access separately for each resource.

OAuth scopes

Scope	Grants
decisions:read	View decisions, approvals, and history
decisions:write	Create, approve, reject, or update decisions
agents:read	View agent work, status, and execution history
agents:write	Delegate, hand off, retry, pause, resume, or cancel work
initiatives:read	View initiatives, work state, artifacts, and receipts
initiatives:write	Create or update work and attach artifacts or proof
memory:read	Search and recall organizational memory
offline_access	Allow the client to refresh access after the current session

OrgX evaluates the required scope when a tool runs. A connection can discover a tool and still receive a structured 403 insufficient_scope response when the OAuth grant is too narrow. Reconnect and authorize the missing resource action; do not retry the same call unchanged.

Interactive widgets and resources

Clients that support MCP Apps can render OrgX views inside the conversation.

Kind	Name or URI	What it returns or opens
Resource	orgx://initiative/	Initiative details with milestones and tasks as markdown
Widget	ui://widget/decisions.html	Pending decisions and approval actions
Widget	ui://widget/agent-status.html	Current agent status and activity
Widget	ui://widget/search-results.html	Search results for organizational memory
Widget	ui://widget/initiative-pulse.html	Initiative health and progress
Widget	ui://widget/task-spawned.html	Confirmation that agent work was assigned
Prompt	create-roadmap	A guided initiative, milestone, and task creation flow
Prompt	plan_feature	A planning flow with improvement suggestions
Prompt	get_my_patterns	Learned planning patterns and skills

A widget is a rendered view of tool output, not a second source of truth. Re-call the underlying read tool when freshness matters—especially after a state change, reconnect, or long-running agent run. Hosts without MCP Apps support still receive the structured tool result.

MCP or REST?

Use MCP when you need	Use REST when you need
A person or assistant working interactively	A server-to-server integration
OAuth, client-managed context, and inline widgets	A versioned contract, webhooks, or batch processing
Natural-language routing across OrgX tools	Explicit idempotency, pagination, and HTTP error handling

The MCP ↔ REST parity page maps the surfaces and documents the canonical v1 routes.

Troubleshooting

The client connects but tools are missing

Call orgx_bootstrap, inspect visible_tools, and confirm the endpoint profile. Reconnect after changing a profile or OAuth grant so the client refreshes its tool list.

A call returns 403 insufficient_scope

The OAuth grant does not include the resource action required by the tool. Reconnect and choose the missing access, or use a read-only workflow.

The workspace or results are wrong

Stop before writing. Call orgx_bootstrap with the intended workspace, then use orgx_search or orgx_inspect to verify the record boundary.

A widget looks stale

Treat the widget as a snapshot. Re-call the corresponding read or status tool and compare the returned state before telling the user that work is complete, blocked, or still running.

An agent is not producing an outcome

Check get_agent_status, inspect the bound task or initiative, and look for a pending decision or blocker. A queued run is not evidence that the final artifact exists.

You need to use the local OpenClaw bridge

Confirm the OpenClaw gateway is running and use the local bridge setup guide. The hosted URL and local bridge have overlapping tools but are not identical transports or surfaces.

Next steps

MCP Client Setup

Configure Claude, Cursor, ChatGPT, VS Code, OpenClaw, or another remote client.

MCP Tools Reference

Browse exact inputs, scopes, profiles, and response examples.

Agent Quickstart

Run a first workflow from context through accountable completion.