Current Proof Run
Wake up to the brief, not the rebuild.
One morning brief compressed the state into the three calls that actually mattered next.
OrgX MCP lets a supported AI client retrieve and work with the initiative records your team keeps in OrgX. The aim is simple: when a conversation ends, its decisions and outputs should still be available to the next session.
This is the March 2026 launch article. The setup links below point to current documentation; later directory listings and client releases have their own dates. A historical demo is not proof of present-day parity across every client.
What the connection adds
MCP is the Model Context Protocol: a way for an AI client to discover and call tools exposed by another service. Connecting OrgX gives the client an authorized route to OrgX records. It does not automatically make every message in every tool part of company memory.
For example, an agent revising an incident runbook needs more than the task title. It needs the chosen recovery procedure, the current runbook, and any unresolved approval. OrgX links those records to the initiative so the agent can inspect them before proposing a change.
The public server repository contains the implementation and setup guidance. The hosted endpoint is:
https://mcp.useorgx.com/mcpUse the client-specific setup guide, because transport, authentication, installed tools, and host capabilities differ. The wizard is another entry point:
npx @useorgx/wizard@latest setupReview what setup will change. Complete the connection flow and approve the workspace access you intend to grant. A local configuration check cannot substitute for checking the tools available inside the authenticated client.
Try one existing decision first
Begin with a read, using work you can recognize:
OrgX, confirm which workspace I am connected to. Find our incidentinitiative and retrieve its latest recovery decision with a sourcelink. Do not create work, start agents, or approve anything.Inspect the result. Does it name the correct workspace? Does the decision belong to that initiative? Can you open its source? An empty workspace may have no matching decision; a connection failure and an empty result need different fixes.
A failed authorization check needs reconnection or access review, not a broader search.
Only then ask for an authorized change. Creating a plan or initiative and launching agent work are separate operations. Use the tools advertised by your installed profile, inspect their results, and keep approvals and execution costs explicit.
What a durable handoff looks like
Here is an illustrative handoff, rather than a customer outcome claim. One agent revises an incident runbook. The owner requires a recovery step to remain manual. Another agent continues the work the following day.
The second agent should retrieve:
| Record | What it answers |
|---|---|
| Initiative | Which incident is this work part of? |
| Decision and source | What scope did the owner choose, and why? |
| Artifact version | Which draft is being changed? |
| Review record | What was accepted, rejected, or left unresolved? |
| Next action | What may happen now, and what still needs permission? |
This illustrative handoff lists the records to inspect; it does not establish automatic capture or recovery.
If it cannot retrieve those records, it should report what is missing. Inventing a recovery procedure or treating the last agent's summary as approval would make the handoff less trustworthy.
The launch widget captures illustrate the workflow, not customer-average performance or a fresh-session recovery test for every client.
How to check whether it worked
Close the first conversation. In a fresh supported session, ask OrgX for the same initiative, decision, artifact version, and unresolved review. Compare the retrieved links and owner ruling with the original records.
That comparison distinguishes continuity from a fluent recap. The agent may recall the incident while missing the decision that keeps a recovery step manual. Check the record that governs the next action, not just whether the summary sounds familiar.
Client write support, automatic capture, and cross-client recovery remain partial beta capabilities. A connected client may expose a narrower tool profile; interactive widgets also depend on the host. Consult the integration guide before moving a real workflow onto a particular client pair.
Start with one existing initiative and one decision. Confirm that the next session can retrieve the evidence before asking it to continue the work.