Start with read-only MCP
If your team needs inspection first, start with the protocol boundary and confirm what agents can read before you talk about packages or skills.
Use this page when the practical question is simple: which entry surface fits our team today, and what should we avoid overclaiming?
In plain language: Codeflow already works with real Codex and Claude Code workflows, can sit next to OpenClaw without fake plugin claims, and still ships a read-only MCP boundary rather than a hosted operator story.
Here, read-only MCP means a protocol layer for inspecting runs, repo-owned skills means reusable playbooks, and builder packages means the thin API and package entrypoints for teams that need integration.
Open integration guide, builder quickstart, AI + MCP + API surfaces, or the ecosystem map.
If your team needs inspection first, start with the protocol boundary and confirm what agents can read before you talk about packages or skills.
If the main problem is repeatable agent behavior, start with the skill pack and reuse the repo's own playbooks instead of inventing plugin language.
If you need a thin client, contract-facing types, or shared frontend substrate, move down the builder ladder after the protocol boundary is clear.
If you need to prove the story to a team first, start with the public proof-first loop, the news_digest baseline, and the share-ready Workflow Case recap before deeper adoption.
Use the detailed matrix once you already know the kind of entry surface you want.
| Surface | Codex | Claude Code | OpenClaw | Truthful boundary |
|---|---|---|---|---|
| Command Tower + Workflow Cases | Primary workflow fit | Primary workflow fit | Adjacent review surface only | Use these as the shared operating record. Do not turn them into hosted-team-account claims. |
| Proof & Replay | Yes | Yes | Yes, if explicitly mapped through the same review loop | Proof and compare are real today. They do not imply a broader plugin marketplace. |
| Read-only MCP | Yes | Yes | Yes, as a read-oriented protocol layer | The shipped public MCP contract is read-only. |
| Repo-owned skills | Direct fit | Direct fit | Adaptation layer | Skills are repo-owned playbooks, not official plugin listings. |
| Builder packages | Thin client + contract + shared substrate | Thin client + contract + shared substrate | Usable when explicitly mapped | Useful package entrypoints today, but not a full SDK platform. |
| Official plugin/store path | Native CLI / IDE surfaces exist; Codeflow is not published as a Codex directory item today | Official plugin surfaces exist; Codeflow is not published there today | Official plugins + ClawHub exist; Codeflow is not published there today | Keep host-platform reality separate from Codeflow publication state and the underlying Codeflow runtime names. |
"Codeflow works with Codex and Claude Code workflows through Workflow Cases, Proof & Replay, read-only MCP, repo-owned skills, and builder entrypoints."
"Codeflow already ships the official Codex, Claude Code, or OpenClaw plugin."
"OpenClaw stays adjacent and can use the same review/proof layer when an explicit mapping is shipped and tested."
"OpenClaw is a first-class native surface inside Codeflow today."
No. This is a truthful compatibility and adoption page for repo-owned surfaces that already exist today. Plugin wording stays narrow and ecosystem-specific instead of becoming the default umbrella term.
No. The public MCP contract remains read-only. Guarded queue and operator mutation groundwork does not change that public promise.
Usually protocol first, then packages. Start with MCP or API if you need the truth boundary, then move to builder packages once the fit is clear.
OpenClaw belongs in the adjacent coding-agent layer unless a real mapped bundle or native path is explicitly shipped and tested.