PAPI + Codex CLI
PAPI runs inside Codex as an MCP server, from one block in config.toml. Fresh sessions open knowing the plan, the decisions, and the direction, not just the code.
Free on three projects · no card · 14-day refund on Pro
How PAPI fits Codex CLI
No new app to open: PAPI runs inside the AI tool you already build in. One TOML block in ~/.codex/config.toml and PAPI's tools ride along in every Codex session. Type orient and Codex answers with the live state of the project.
Decisions on the record: every call your project makes is kept, with the why, and checked against new work. Codex reads them back before related work, so a long run doesn't reopen a question you settled last week.
Catches drift: the moment new work slides off the plan, PAPI flags it. Codex builds from PAPI handoffs: goal, scope boundary, acceptance checks. When the diff slides past the boundary, the review gate says so before it ships.
Connect PAPI to Codex CLI
- Create your account. Sign in at getpapi.ai with GitHub or email and name your first project.
- Add the TOML block. Append this to ~/.codex/config.toml. Codex opens a browser tab to sign you in on first connect.
[mcp_servers.papi] url = "https://mcp.getpapi.ai/mcp"
- Run setup. In Codex, run setup. PAPI writes your product brief, then plan starts your first cycle.
Questions
- Can't I just use Linear or Jira?
- They're built for teams of humans, and they sit in a tab your AI can't see. PAPI runs inside the tool you build in, so the plan and the work stay in one conversation. Different category; keep Linear if your team lives there.
- Does PAPI read my code?
- Your code stays on your machine. PAPI adds a planning layer to whatever you're already building, it doesn't read your source.
- I use Codex and Claude Code side by side.
- Good, that is the point. PAPI is one remote MCP server, so both connect to the same project and hand off through the same plan. A handoff written in one session executes cold in the other.
- My AI already has memory. Why add PAPI?
- Memory recalls what happened. It doesn't hold a plan, sequence tasks, gate builds, or tell you when the work slid off course. A memory of everything you did is not a plan for what's next.