Atlassian Code Context CLI Guide for Codex, Claude Code, and Cursor

Quick answer

Atlassian exposes Code Context through the Teamwork Graph (TWG) CLI so coding agents can retrieve source and organizational context beyond the local workspace. Atlassian’s launch names Codex, Claude Code, and Cursor as examples. The current support documentation specifically warns that when TWG CLI is used with a third-party agent, retrieved source code is sent to that provider and governed by its terms.

Treat TWG CLI as a context channel, not an action authorization layer. It can help an agent find related code; it does not decide which repository may be edited, which commands may run, or which change may ship.

The control path

developer identity
  -> Atlassian / Teamwork Graph sign-in
  -> query-time SCM permission check
  -> lexical or semantic retrieval
  -> TWG CLI
  -> selected coding agent and model
  -> local files, tools, commands, and network
  -> review, tests, pull request, and human merge

Approve each arrow separately. A user may be allowed to retrieve code but not to send it to a particular model, edit the related repository, run a deployment command, or publish a result.

Codex, Claude Code, and Cursor checklist

Boundary	Record before the pilot
Agent	Exact product, client version, publisher, account, and model
Source access	Atlassian identity, SCM identity, repositories, branches, and groups
Data transfer	Retrieved snippets, prompts, attachments, logs, retention, region, and provider terms
Local authority	Workspace roots, read/write files, shell commands, browser, MCP servers, and network
Cost	Atlassian plan, model/API billing, capacity, token budget, time limit, and owner
Change control	Branch, protected paths, tests, review, CI, environment approval, and merge owner
Recovery	Stop, revoke, rotate, remove logs, revert change, and disconnect CLI/provider

Use only an officially supported integration path for the selected agent. A vendor appearing in a launch example does not guarantee that every client version, authentication mode, extension, or deployment surface is supported.

A safe first task

Choose a read-only, cross-repository question with a known answer, such as locating the implementation and callers of a non-sensitive internal API. Ask the agent to cite repository, path, symbol, and revision, then verify each result in the source SCM.

Next, test failure cases:

• A repository the user cannot access.
• A branch with restricted access.
• A recent push that may not be indexed yet.
• An ambiguous semantic query with plausible but wrong results.
• A stale comment or documentation string.
• A connector or SCM outage.
• A prompt that asks the agent to write outside the approved repository.

Do not begin with production credentials, incident response, vulnerability adjudication, or an autonomous write. Atlassian says semantic search is not a security scanner and AI answers remain probabilistic.

Third-party provider boundary

Atlassian states that source retrieved through TWG CLI may be sent to third-party products and that customers are responsible for assessing provider security and suitability, costs, API capacity, and agent actions.

Before use, confirm provider identity, account type, model, data-use terms, retention, training controls, region, logs, subprocessors, and deletion. Also decide whether the agent may use other tools after retrieval; context access should not silently expand shell, network, repository, issue, or deployment authority.

Read the Code Context security and indexing guide for the Atlassian-side data boundary.

Benchmark the context, not the brand

Atlassian’s local benchmark compares the same agent, model, and question with two context arms: standard Atlassian access through Rovo MCP, and TWG CLI context. Reviewers see blinded Answer A and Answer B, then the report separates answer preference, token use, and time.

Use representative questions and keep the control and test conditions fixed. Do not claim that lower token use proves a better answer, or that one successful query proves cross-repository correctness. Atlassian calls benchmark results directional.

See the Teamwork Graph benchmark documentation and Code Context readiness check.

Frequently asked questions

Does Code Context give Codex or Claude Code write access?

No. Code Context provides retrieval. Write and command permissions come from the local agent environment, repository credentials, tools, and approval policy.

Can Cursor use Code Context?

Atlassian’s launch names Cursor among coding-agent examples. Verify the current TWG CLI documentation and the exact Cursor integration path before enabling source transfer.

Are retrieved snippets kept inside Atlassian?

Not when a third-party agent receives them. Atlassian warns that source sent through TWG CLI to third-party products is governed by those providers’ terms.

Do better search results guarantee safer changes?

No. Retrieval can reduce missing context, but a coding agent can still misunderstand intent, choose the wrong edit, expose data, or run an unsafe action. Preserve tests, review, and human merge authority.

Official sources

• Atlassian: introducing Code Context
• Atlassian Support: what is Code Context
• Atlassian Support: set up Code Context
• Atlassian Developer: Teamwork Graph benchmark

Source check: August 27, 2026. Recheck CLI support, agent integrations, authentication, provider data terms, costs, capacity, actions, and benchmark tooling.