A bug report says “checkout failed again,” but the useful evidence lives outside the repository. Connect Pathbound and Codex can read the customer’s recent events, support history, and account record while it investigates the code.
Pathbound gives Codex a remote MCP tool surface beside the repository. The customer record stays in Pathbound; Codex pulls the specific profile, event, or timeline it needs for the task.
Customer evidence beside the code: recent events, support conversations, email history, company context, and the resolved contact record.
One shared Codex MCP configuration for the desktop app, CLI, and IDE extension on the same host.
Server-side filters and aggregation, so Codex can inspect the relevant event window without loading a raw export into the context window.
OAuth for an engineer working interactively. A scoped REST API key remains available for CI or another headless runner.
Codex stores MCP servers in its configuration and shares them across the desktop app, CLI, and IDE extension on the same host. Add Pathbound once, then complete OAuth in the client you use.
Register Pathbound as a streamable HTTP MCP server:
Run codex mcp login pathbound to open the Pathbound OAuth flow in your browser.
Run codex mcp list to confirm that Pathbound is enabled.
Start Codex and use /mcp to inspect the active server and tools.
codex mcp add pathbound --url https://mcp.pathbound.ai/mcpOpen Settings → MCP servers and select Add server.
Name it Pathbound, choose Streamable HTTP, and enter https://mcp.pathbound.ai/mcp.
Save, restart the app or extension, and select Authenticate beside Pathbound.
Sign in through the browser. The same host configuration is now available to the other local Codex clients.
Once connected, these are real jobs Codex runs against your customer data. Each recipe is one tested prompt—paste it and go.
When a support ticket comes in, look the customer up in Pathbound before drafting a reply. Pull their unified profile, read the prior conversations we’ve had with them across every channel, and check what they did in the product and on the site leading up to this. Summarize it in three lines—who they are and their plan, the relevant history, what they most likely need—then draft a reply that already accounts for all of it, so I never have to ask them to explain again.See the full recipe—Full-context support
More jobs in use cases, each with a tested prompt.
No. OpenAI’s Codex MCP documentation says those local clients share MCP configuration on the same host. Configure Pathbound once, then restart or refresh the client you open next.
Codex uses ~/.codex/config.toml by default. You can also scope an MCP server to a trusted project with .codex/config.toml. The CLI command writes the configuration for you.
After adding the server, run codex mcp login pathbound or select Authenticate in the desktop or IDE server list. Both open Pathbound’s OAuth flow in your browser.
No. ChatGPT web installs Pathbound as a developer-mode app under Plugins. Codex clients store Pathbound as an MCP server in Codex configuration. See the ChatGPT setup for the web path.
Yes, for a headless process that cannot open a browser. Create a Pathbound REST API key with only the scopes the process needs and pass it as a bearer token. Interactive Codex users should normally use OAuth.
Only through write tools allowed by the Pathbound credential and connected source. Start read-only for debugging and investigation. Enable individual write paths only when the workflow requires them; versioning and rollback are not live yet.
Sign up, connect one CRM or drop the events snippet, and point your MCP client at mcp.pathbound.ai/mcp.