CodexPro MCP Server: Give ChatGPT Local Coding Tools

• One entry point that runs the other CodexPro actions

  A single tool that lists and runs the other CodexPro tools by action name, such as config, open, changes or handoff.

  Try: “List the CodexPro actions you can run.”

  Read · codexpro

• Shows how CodexPro is set up

  Shows the safety modes, limits and blocked paths. It does not reveal the access token.

  Try: “Show me the CodexPro settings and which paths are blocked.”

  Read · server_config

• Runs a local self-check

  Runs one local diagnostic of modes, workspace access, git and the safe command policy. It may write a probe file under .ai-bridge but does not touch source files.

  Try: “Run the CodexPro self test.”

  Files · codexpro_self_test

• Opens the repo CodexPro started in

  Selects the default repo for this chat, so later calls can leave out the workspace.

  Try: “Open the current workspace.”

  Read · open_current_workspace

• Switches to another allowed repo

  Opens and selects one of the repos you allowed when you started CodexPro.

  Try: “Open my api repo.”

  Read · open_workspace

• Maps the repo

  Builds a bounded map of languages, project types, entry points, areas, symbols and how parts relate, with coverage warnings.

  Try: “Inspect the workspace and tell me how it is organized.”

  Files · inspect_workspace

• Lists files and folders

  Shows the files and directories in the repo, leaving out blocked paths.

  Try: “Show the folder structure of src.”

  Files · tree

• Finds code in the repo

  Searches by text, pattern, symbol, references or likely impact.

  Try: “Find every place that calls createInvoice.”

  Files · search

• Reads one file

  Returns a text file with line numbers.

  Try: “Read src/auth/login.ts.”

  Files · read

• Looks at an image in the repo

  Returns a PNG, JPEG, GIF or WebP image from the workspace with its size and hash.

  Try: “Look at docs/diagram.png.”

  Files · view_image

• Creates or overwrites a file

  Writes a text file inside the repo and returns a diff. Only available in workspace write mode, which is the default.

  Try: “Create a README for the utils folder.”

  Write · write

• Replaces exact text in a file

  Makes a targeted text replacement and returns a diff.

  Try: “Rename the function getUser to fetchUser in this file.”

  Write · edit

• Applies a patch across files

  Applies one unified diff after checking the paths.

  Try: “Apply this patch to fix the date parsing.”

  Write · apply_patch

• Brings a ChatGPT attachment into the repo

  Saves a file you attached in ChatGPT into the workspace. It accepts only ChatGPT file objects, not arbitrary links, and does not overwrite by default.

  Try: “Import the attached spec into docs.”

  Write · import_file

• Runs an approved check command

  Runs one allowlisted command in the repo, such as tests, build, lint or typecheck. Safe mode can still run your repo's package scripts, so turn bash off for untrusted code.

  Try: “Run the tests.”

  Runs code · bash

• Summarizes your current changes for review

  Gives git status, diff stats and optional diff, plus affected areas and related tests.

  Try: “Review what I changed.”

  Files · show_changes

• Loads a skill from the workspace or your plugins

  Reads the body of a skill file found in the workspace, user or plugin folders, by name.

  Try: “Load the release-notes skill.”

  Files · load_skill

• Reads the planning files used for handoff

  Reads the shared .ai-bridge planning files used to pass work between ChatGPT and a local agent.

  Try: “Read the current handoff plan.”

  Files · read_handoff

• Waits for a local agent to report

  Checks the handoff run state and returns status, diff, log and test excerpts. It only reads.

  Try: “Check whether the handoff run has finished.”

  Files · wait_for_handoff

• Writes a context file for planning

  Creates .ai-bridge/pro-context.md with the repo tree, git state, chosen files and handoff context, for chats that cannot call tools.

  Try: “Export a context bundle for planning.”

  Write · export_pro_context

• Writes a plan for a local coding agent

  Writes .ai-bridge/current-plan.md for Codex, OpenCode or another agent. It only writes files and does not run the agent.

  Try: “Write a plan for Codex to implement the login change.”

  Write · handoff_to_agent

Watch out for

• Your machine is reachable over a public tunnel

  Anyone who gets the tunnel URL with its token can use the tools. Treat the URL like a password, prefer a bearer header, and stop CodexPro when you are done. github.com

• It can edit files and run project commands

  Write mode is on by default, and bash runs allowlisted checks that can still run your package scripts. Use --no-bash and a read-only or handoff mode for untrusted repos. github.com

• The token can end up in a URL

  The ?codexpro_token= form can appear in browser history or logs. The README prefers an Authorization header when the client supports one. github.com

• Blocked paths limit what it can reach

  The README says paths such as .env, keys and .git are blocked, and writes outside allowed roots are rejected. Still keep secrets out of the repos you allow. github.com

Security advisories