Choosing between Teamwork Graph CLI and Rovo MCP
What is this topic about?
Teamwork Graph (TWG) CLI and Rovo MCP are the two official ways for AI agents to work with Atlassian data via the Teamwork Graph.
They target different runtimes, authentication models, and workflow shapes. Use this page to decide which one to start with for your agents and automation.
When to use this topic
Use this topic when you:
Are designing or configuring AI agents that need to access Atlassian data.
Need to choose between a CLI-based approach and an MCP-based approach.
If you need detailed setup or command documentation, see the related topics instead.
概要
Teamwork Graph (TWG) CLI and Rovo MCP are both official ways for AI agents to work with Atlassian data via the Teamwork Graph.
They are complementary, not interchangeable:
TWG CLI is a command-line, agent-oriented interface (
twg) best suited to shell, CI/CD, and deep, graph-style automation.Rovo MCP is an MCP server that exposes tools to MCP-capable hosts (web LLMs, IDEs, or sandboxes) using OAuth and host-native auth flows.
Both surfaces are actively evolving. Revisit your choice as coverage, orchestration, and host capabilities change.
Decision guide
If you… | Lean toward |
|---|---|
Run agents in a terminal, CI/CD, or need scriptable | TWG CLI |
Run agents in web LLMs, IDEs, or sandboxes without a reliable shell / binary | Rovo MCP |
Need OAuth and host-native MCP auth patterns today | Rovo MCP |
Need very broad Atlassian / graph-style command depth (validate for your entities) | TWG CLI (check coverage for your case) |
Want fewer turns for straightforward tasks via pre-declared, parallel tools | Rovo MCP |
At a glance
| TWG CLI | Rovo MCP |
|---|---|---|
What it is | Command-line, agent-oriented | MCP server exposing tools to MCP-capable LLM hosts |
Best for | Shell, automation, deep/broad graph-style workflows | Web LLMs, IDEs, OAuth, sandboxed hosts |
coverage | Generally larger Atlassian / graph-oriented surface — verify your scenario | Curated Atlassian tools; can extend to non-Atlassian systems where configured |
Discovery / turns | Help, flags, examples → can be slower first run, then repeatable | Tools pre-declared; often fewer turns for simpler tasks |
Tokens | Often more efficient in some flows (no huge catalog every prompt) — not universal | Depends on host + manifest; roadmap may shift trade-offs |
Auth (today) | PAT / API token; OAuth planned | OAuth + tokens (host-dependent) |
ランタイム | Not available everywhere (e.g. some sandboxes cannot run a CLI) | Works where MCP is supported |
One org, one session, many tools
Across the organisation you can use both TWG CLI and Rovo MCP (different teams or environments).
Within one agent runtime or session: pick one primary surface (TWG CLI or MCP), or always name the surface each time (“Use MCP” / “Use TWG CLI”). Put defaults in
agents.md,claude.md, or your team agent config so behavior is predictable.TWG CLI + Rovo MCP + RovoDev CLI (or other overlapping sources) in the same profile: prefer one primary unless you have a clear reason not to; if several are installed, instruct explicitly which to use, or use separate profiles (e.g. MCP-only vs CLI-only) or a documented trigger convention.
When to use which
Use TWG CLI when
You need broad coverage, multi-hop or fine-grained queries across products (after you validate CLI supports your entities/actions).
The agent runs in a terminal, CI/CD, or other shell-capable environment.
You want to chain to local tools, persist JSON, or wire automation pipelines.
Use Rovo MCP when
You integrate web-based LLMs, IDEs, or third-party sandboxes without shell access.
You want fewer tool turns, pre-declared tools, and parallelizable orchestration for simpler tasks.
You need OAuth-based access and governance patterns your MCP host supports well.
FAQ
Is TWG CLI basically the same as Rovo MCP?
No. CLI = shell commands and scripting; MCP = tools exposed via an MCP client directly to the LLM. Complementary, not interchangeable.
We already use Rovo MCP. What changes if we add TWG CLI?
You add a command-line path for places that can run twg (scripts, automation, pipelines). Everything you already do through MCP stays; this is extra for environments that fit CLI better.
Can we use both in the same org?
Yes — see One org, one session, many tools.
Is one always faster or “better”?
No — it depends on task complexity, tool exposure, planning, and product investment. Benchmark your own workflows if needed.
Is TWG CLI offline?
No. It is a local binary that accesses cloud data/services where required.
TWG CLI + Rovo MCP + RovoDev CLI in one environment?
この内容はお役に立ちましたか?