Skip to main content
Glama

Convorel 1.0

Convorel lets a coding agent hold review conversations with ChatGPT through your logged-in Google Chrome, and share an explicitly allowed workspace through a guarded MCP server. Your agent owns edits, tests, and decisions; ChatGPT provides a second opinion.

The Go executable contains a thin JSON CLI, a resident daemon, direct CDP browser control, and an MCP stdio server. Linux x64 and arm64 releases are static binaries. Chrome and a ChatGPT account with access to your selected model are required for conversations. Workspace Git queries require Git. MCP connectivity additionally requires an OpenAI Secure MCP Tunnel client and a configured ChatGPT developer app.

Install

curl -fsSL https://raw.githubusercontent.com/MarioJames/convorel/main/install.sh | bash
convorel setup --workspace /absolute/project
convorel service install
convorel login

Sign in yourself in the dedicated Chrome window. Configuration is private; workspace sharing is a separate explicit choice. See quickstart for tunnel configuration and security before granting access.

Related MCP server: WorkspaceGuard MCP

Ask and continue

convorel ask --id design-review --workspace /absolute/project \
  --type DES --topic 'Storage boundaries' --prompt-stdin <<'PROMPT'
Review the storage design against the documented invariants. Identify blocking risks with file references.
PROMPT
convorel wait --id design-review
convorel status --id design-review
convorel followup --id design-review --request-id revision-2 \
  --prompt 'Review the revised implementation and validation evidence.'
convorel wait --id design-review
convorel finish --id design-review

Real reviews default to Pro (effort 5). Tests and acceptance use Instant (--effort 1). Save the returned run and use --run when waiting for a specific round. wait returns the completed Markdown immediately; background naming is reported separately by status. A killed wait process leaves the daemon observing. Uncertain delivery is never automatically resent.

Each command prints one JSON document and uses exit 0 for success, 1 for errors, and 2 for pending or uncertain outcomes. MCP stdio is the protocol-only exception. See usage, the CLI contract, and the bundled review skill.

Upgrade and data

convorel upgrade
convorel skills install --target /absolute/skills/chatgpt-review

The installer verifies SHA-256 and retains the active and previous release. Modified or unknown installation files survive cleanup. Skill installation preserves customizations by refusing an unmanaged or modified target. Stop all 0.9 processes before switching: 1.0 reuses tasks.db, conversations.db, and diagnostics.db in place, with no legacy command aliases. Preferences stay in ~/.config/convorel/preferences.json; operational config and databases stay in ~/.local/share/convorel.

Develop

Use Go 1.27.4 with CGO disabled for release builds:

go build -o /tmp/convorel ./cmd/convorel
go vet ./...
go test ./...
scripts/build.sh /tmp/convorel-release
scripts/test-install.sh /tmp/convorel-release

Headless tests use disposable Chrome profiles and local fixtures; they do not contact ChatGPT. Optional workspace sandbox tests require Bun, Bubblewrap, and prlimit because the retained remote executor supports Bun projects. Convorel itself needs no Bun or Node runtime. See contributing, architecture, and validation.

This independent community project is not an official OpenAI product. See LICENSE, NOTICE, and third-party notices.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      D
      maintenance
      A read-only MCP server that lets AI assistants inspect local workspace files, search context, view git diffs, and fetch a fixed GitHub profile, all within a sandboxed stdio transport.
      5
      ISC
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables ChatGPT or any MCP client to operate safely on a designated workspace by listing, reading, searching, writing, and trashing files, inspecting Git status/log/diff, and optionally running allowlisted executables without a shell.
      Apache 2.0
    • A
      license
      A
      quality
      C
      maintenance
      Enables ChatGPT Web to inspect a single local workspace read-only via the OpenAI Secure MCP Tunnel, with tools for listing files, reading text, searching, and viewing Git status/diff.
      6
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables MCP clients and AI agents to search, read, list, and reindex locally indexed private code repositories using read-only, path-confined tools.
      MIT