How to integrate Unify with Codex
To integrate Unify with Codex, install the Unify marketplace plugin in the Codex app and connect the intended Unify account. In a supported task view, select it through Sources and Use plugins. Use the connection to bring Unify research or record context into an operations task, then inspect the Agent result before using it in an artifact.[1][3]
- Connection
- Official MCP
- AI assistants
- Tool
- Codex
- Unify-documented client setup
- Data flow
- Codex ↔ Unify
- Tool calls and returned results
At a glance
- Use the Unify-documented Codex setup
- Use tool calls and returned results instead of an outbound Play payload
- Check the final Agent result before treating the task as complete
When this workflow makes sense
Start with the decision your team needs to make, then choose the connection that supports it.
Codex can help a technical operations owner turn a business question into a reusable deliverable. The useful boundary is between data requested from Unify and local work built from that data. For example, an account brief can inform a review document without modifying production records. Keep the source run and its assumptions visible so a polished artifact does not hide incomplete research.
A good fit when
Technical RevOps teams using Codex tasks to produce account reviews, operational documents, or code informed by Unify context.
Reconsider the plan when
This page covers the Codex app marketplace path documented by Unify. Do not assume the same installation steps apply to every Codex CLI or IDE surface.
Design the handoff
Use this framework to agree on the input, the next action, and who owns the result.
Suggested workflow design, based on the documented capabilities below.
| Stage | Decision to make |
|---|---|
| Client and account | A Codex app workspace that permits the published Unify plugin and its required app connection |
| First task | Suggested prompt: “Use Unify to research example.com. Create a local account-review draft with sources, open questions, and the Unify run reference. Do not update the CRM.” |
| Execution | Define whether the request is research-only or includes a specific authorized action. Keep the run ID for follow-up. |
| Result | Review the final output and unresolved questions before using the result in a sales decision. |
Explore the scenarios
Choose the situation that fits your team and work through the proposed plan.
Open a scenario to explore its plan. Checklists track your selections in this page only.
Build an account-review artifact
An operations owner needs a reusable brief built from current Unify research.
Preserve the source run and unresolved questions alongside the artifact.
- Select Unify for the Codex task
- Request a bounded account research run
- Create the review artifact only after reading the completed result
Separate local edits from CRM changes
A task mixes editing an internal document with a potential account update.
A request to draft a recommendation is not a request to apply it to the CRM.
- State which outputs are local and which would change Unify or the CRM
- Review the proposed field changes and record identities
- Request the specific remote update separately if it is wanted
Illustrative workflow, not a preconfigured automation. Complete these checks in your own account.
Define success before launch
Inspect the data and the intended outcome separately. Connection status is only the first check.
Expected, not tested
After the connection is authorized, Codex can identify the intended Unify user and retrieve the completed output of the controlled research task.
Inspect get_me, the requested task, the final run status, and read_agent_results together. A discovered tool or a run ID confirms a different stage from a completed result. For any explicitly requested write, inspect the actual destination record as well.
Documentation-based reference. This workflow has not been tested end to end. Use a record or event you control before enabling live activity.
Questions answered
Common decisions when connecting Unify and Codex.
How do I connect Unify to Codex?
In the Codex app, open Plugins, search for Unify, and install its marketplace plugin. Check any workspace restrictions and required app access shown on the listing.[1][3]
Where do I select the Unify plugin in a Codex task?
In a supported Codex task view, open Sources, choose Use plugins, and select the installed Unify plugin. Available controls and access depend on the workspace and product surface.[1][2][3]
How is MCP different from a Play webhook?
An MCP client can invoke Unify tools and retrieve their documented results. The Play Webhook action sends an external request and does not pass its response into downstream Play steps. They are separate connection mechanisms.[2][5]
Does a run ID mean the task succeeded?
No. run_agent starts background work. Check the run with poll_agent, handle any clarification, and call read_agent_results after the run reaches READY or ERROR.[2]
Limits & decisions
Check these boundaries before designing a live workflow around the connection.
- This page covers the Codex app marketplace path documented by Unify. Do not assume the same installation steps apply to every Codex CLI or IDE surface.[1][3]
- Unify permissions and Agent usage limits still apply[1][2]
- A completed research result does not itself authorize sending, enrollment, or CRM updates
- No authenticated end-to-end execution has been performed
What the connection supports
Understand the connection, the data involved, and the behavior that matters for a Unify + Codex workflow.
The assistant requests work from Unify and retrieves its result using the connected user’s permissions.
| Decision | What to know |
|---|---|
| Connection route | Unify explicitly documents Codex setup[1] |
| Client requirement | A Codex app workspace that permits the published Unify plugin and its required app connection[1][3] |
| Published server | https://mcp.unifygtm.com/mcp (authentication required)[4] |
| Identity | Calls are scoped to the authenticated Unify user and tenant[2] |
| Agent lifecycle | Start a run, poll status, answer clarification, and read the terminal result[2] |
| Usage | Unify MCP Agent work uses existing Unify Chat usage and credits[1] |
MCP exposes tool schemas to the client. For Agent work, provide a natural-language prompt and retain the returned runId for status and result calls. Direct API tools have their own required inputs. This connection is separate from a Play’s outbound Webhook action and does not make webhook responses available to later Play steps.
Technical setup reference
References and requirements for the Codex connection.
Connection requirements and setup checklist
Before connecting
- A Codex app workspace that permits the published Unify plugin and its required app connection
- A Unify user with access to the intended workspace and available Chat credits for Agent runs
- One controlled research task with a clear output and explicit boundaries on record changes
Attach Unify to the task
Open Sources, select Use plugins, and choose Unify where those controls are available. State the local deliverable and whether remote records may change before requesting work.[3]
Check the Agent result
For Agent work, run_agent returns a runId before the task finishes. Poll that run. Answer every returned question if its status is CLARIFICATION_NEEDED. On READY, read_agent_results returns the completed output; on ERROR, inspect the reported failure.[2]
Troubleshooting
Start with the stage where the expected data or action stops appearing.
The client cannot find or authorize Unify
Check the official installation path for this client, workspace policy, and required account connection. Restart clients where Unify’s setup calls for it, then repeat sign-in.[1][3]
A run starts but no result appears
Keep the original runId and poll its status. PENDING is still running; CLARIFICATION_NEEDED requires answers; READY permits reading the result. Use list_agent_runs to find a previous run rather than starting duplicate work.[2]
The tool reports missing access or unavailable credits
Verify the connected user and workspace. Unify scopes tools to that user, and Agent runs require available Chat funding and are subject to rate limits. A table owned by another user is not accessible through load_datatable.[2]
Sources & review
Official documentation reviewed on September 16, 2026.
Technical claims link to primary documentation from Unify and the destination tool. Scenarios, field choices, and recovery plans are editorial implementation examples. Documentation review is separate from execution testing.
Documentation reviewed; end-to-end execution not tested. No named technical reviewer has signed off on these articles yet.
Content updated: . Report a correction to Unify with the tool name, source URL, and behavior you observed.
Keep exploring
