More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    peek

    1

    Inspect Claude Code and Codex session turns, plans, diffs, and telemetry via MCP.

    Free

    1 useSecurity scanned
    peek

    peek

    Example session with this skill installed

    /peek status Peek logo design

    • Read your context and instructions
    • Compiled the peek

    Peeked at session Peek - Peek logo design 6987b1cf (status only — live diff present, no plan).

    Status

    Turns9
    Time110s wall · 110s active (no idle)
    Cost$2.21 (telemetry: receiving)
    Tokens418 in · 9,586 out · 1.38M cache read · 69k cache write
    Permissions33 hook-allowed · 1 denied (Artifact, blocked by hook) · 0 user prompts
    ModeacceptEdits since session start

    Touched files: 8 — the logo SVG set (peek-mark, peek-wordmark, peek-lockup, light + dark variants) and preview.html under ~/GolandProjects/peek-mcp/.../docs/assets/logo/.

    The session is active and healthy: all edit traffic on the logo assets, one tool denial worth a glance, nothing idle. For the actual changes: ask for the diff of this session — it's live.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Every Claude Code and Codex session on your machine, fully observable — from whatever agent you're asking in.

    What you see

    Peek is the complete breakdown of what an agent actually did:

    • Turns — with the agent's reasoning, not just the messages.
    • Tool calls — every call with its input, error state and timestamp.
    • Subagents — each spawned agent as its own scopeable transcript.
    • Events — the typed stream: which skills fired, which permissions were denied, when the mode changed.
    • Plan — the agent's current task list and state.
    • Diffs — the merge-base diff and the live uncommitted diff.
    • Memory — the session's auto-memory.
    • Numbers — input/output and cache tokens, cost, wall and active time.
    • Telemetry — with Claude Code's telemetry enabled, OTLP metrics and tool-detail logs ride along.

    How you use it

    Ask "what is Claude Code doing?" for the latest session. Or pick a view:

    AskReturns
    peek listevery session, with plan/diff flags
    peek / peek 10recent turns
    peek plan / peek diffjust that section
    peek statusbounded counters — cost, wall/active time, no turns
    peek <id or title>a specific session, claude or codex

    status exists for monitoring loops: full-turn polling kills sessions with prompt-too-long — bounded counters don't. Pagination is handled; large sessions come in pages, not truncations.

    Setup: none

    Peek rides on peek-mcp, a small open-source (MIT) Go server. The skill detects a missing server, installs it, registers the full configuration for your agent, verifies the spawn, and tells you exactly when the tools appear — immediately, after a reconnect, or next session, depending on your agent. No sessions to read in this environment? It says so plainly instead of guessing.

    Local only

    A stdio server reading ~/.claude/projects/ and ~/.codex/sessions/. Your transcripts never leave your machine; the skill adds no network path.


    Tested per version — smoke-checked and A/B-measured before every release. Source and version history: peek-mcp.

    Works only with local session stores (~/.claude/ or ~/.codex/). CLI tools like Claude Code require a manual /mcp reconnect after installation.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 18 days ago

    • Free with an Agensi account

    Requires

    Peek MCP
    Claude Code or Codex
    macOS

    Permissions

    Write Files
    Terminal / Shell
    Read Files
    Network Access

    File Scopes

    ~/.claude/projects/**
    ~/.codex/sessions/**
    ~/.cursor/mcp.json
    ~/.codex/config.toml
    ~/.claude.json

    peek-mcp instalation points codex and claude session sources

    Listed25 days ago
    Updated18 days ago

    What's inside

    Frequently Asked Questions

    Peek provides deep visibility into the "black box" of agent sessions, allowing you to inspect internal plans, pending git diffs, and session turns that are usually hidden in local logs or complex JSON files.

    This skill is compatible with Claude Code, Codex CLI and Cursor; it requires a local session store at either `~/.claude/projects/` or `~/.codex/sessions/`.

    Installation is automated via a built-in bootstrap process that installs and registers the `peek-mcp` binary, though you will need Go installed on your system to complete this setup.

    Peek helps to reduce token usage, as it is a fantastic tool for session handoff - giving a fresh session enough context (but not too much) to prevent further cache read cost. Its great!

    You receive the full MCP skill package, including the `peek-mcp` server and the automated environment bootstrapping tool for seamless integration into your existing workflow.

    No. peek-mcp is a local stdio MCP server. We require network access only for the potential one time installation of peek.

    No — the skill detects a missing server, installs it, registers the full configuration for your agent and verifies the spawn. In Claude Code, run /mcp and reconnect; Cursor picks the config up live; Codex CLI loads it next session.

    Yes — any MCP-capable agent on the machine can be the viewer; Claude Code and Codex are the sources.

    New in DevOps & Cloud