Developers · Codex CLI capture

Capture from Codex CLI.

5 minutes. For terminal-driven workflows: infrastructure work, debugging sessions, CLI exploration.

Step 1, create an MCP token

Generate one at Settings → Integrations → MCP tokens. Copy it (shown once).

Step 2, add Pulse to Codex config

Codex uses TOML (not JSON). Edit ~/.codex/config.toml:

[mcp_servers.pulse]
url = "https://pulsehq.tech/api/mcp"

[mcp_servers.pulse.http_headers]
Authorization = "Bearer pulse_mcp_PASTE_YOUR_TOKEN_HERE"

TOML pitfalls

No trailing commas (unlike JSON).
String values are double-quoted.
The table name is [mcp_servers.pulse], underscore, not [mcp.servers.pulse].
Sub-tables come after the parent table.

Step 3, verify (no restart needed)

Unlike Claude Desktop and Cursor, Codex picks up MCP config changes on the next invocation. Just run codexin your terminal. Sanity check: “Remember that staging mirrors prod nightly” should call pulse_remember, a capture-only token can write and recall its own memory, but read tools like get_pulse need a pk_live_ read key.

Step 4, your first capture

Run codex, work through a problem, then say:

Save this debugging session to Pulse, tag it deploy-failures.

Codex captures naturally include terminal output and shell commands. Pulse preserves the full context. For structured decisions: “We’re using Cloudflare R2 for storage. Alternatives: S3, GCS, Backblaze B2. Save the decision to Pulse.”

Power user tips

Shell aliases help: alias save="codex save the last session to Pulse".
Long sessions: capture the conclusion as a decision rather than the full transcript.
Bouncing between projects? Label tokens by project ("Codex, project-trellis") so you can revoke one without affecting the others.