Skip to content

chatgpt-app-setup.md

Latest commit

 

History

History
88 lines (72 loc) · 3.47 KB

chatgpt-app-setup.md

File metadata and controls

88 lines (72 loc) · 3.47 KB
Raw

ChatGPT Custom MCP App Setup

ChatGPT product proof is manual and workspace-gated. This repository can prepare the server, OAuth/dev-noauth config, MCP metadata, tool list, and activation package, but it cannot claim product UI proof until an operator exercises ChatGPT.

Server Preconditions

./bin/codencer setup self-host \
  --gateway-url http://127.0.0.1:19090 \
  --mcp-url http://127.0.0.1:19090/mcp \
  --relay-url http://127.0.0.1:8090 \
  --relay-request-timeout-seconds 300 \
  --default-relay-token-env CODENCER_DEFAULT_RELAY_TOKEN \
  --token-env CODENCER_GATEWAY_MCP_TOKEN \
  --enable-oauth-dev \
  --json

./bin/codencer login --gateway http://127.0.0.1:19090
./bin/codencer connector login --gateway http://127.0.0.1:19090 --relay default --json

./bin/codencer activation self-host \
  --gateway http://127.0.0.1:19090 \
  --relay http://127.0.0.1:8090 \
  --project codencer \
  --token-env CODENCER_GATEWAY_MCP_TOKEN \
  --json

Save the generated client secret and operator approval code from $CODENCER_HOME/tokens.

codencer activation self-host --json writes a package containing ChatGPT setup that points to Gateway. codencer activation chatgpt --json remains available for direct Relay debug setup. The generated setup sheet includes:

  • mcp_endpoint
  • auth_mode
  • client_id
  • client_secret_file
  • operator_code_file
  • authorization_server_metadata
  • openid_configuration
  • authorization_endpoint
  • token_endpoint
  • protected_resource_metadata
  • scopes
  • expected_tools
  • chatgpt_ui_steps
  • test_prompts
  • evidence_checklist

Literal secrets are redacted. File paths are shown only for local token files.

App Values

  • MCP endpoint: http://127.0.0.1:19090/mcp
  • OAuth authorization server metadata: http://127.0.0.1:19090/.well-known/oauth-authorization-server
  • OpenID configuration: http://127.0.0.1:19090/.well-known/openid-configuration
  • Protected resource metadata: http://127.0.0.1:19090/.well-known/oauth-protected-resource/mcp
  • Backend Relay profile target: http://127.0.0.1:8090
  • Required behavior: use codencer.list_projects first.
  • OAuth dev redirect behavior: valid redirect URIs are accepted for self-host dev testing. Production must use redirect allowlisting or an external IdP. OAuth dev mode is not public multi-user production.

Expected project-aware tools include:

  • codencer.list_relays
  • codencer.get_relay
  • codencer.list_projects
  • codencer.get_project
  • codencer.list_project_locations
  • codencer.submit_project_task_and_wait
  • codencer.run_project_manifest
  • codencer.get_run_report
  • codencer.get_blocker

Operator Steps

  1. Confirm the ChatGPT workspace has Developer Mode/custom MCP app access.
  2. Create a draft custom MCP app named Codencer.
  3. Use the MCP endpoint above.
  4. Use OAuth mode against the Codencer OAuth dev metadata for self-host testing, or an operator-owned OAuth front door for production IAM.
  5. Complete the authorization flow with the operator approval code.
  6. Scan/list tools and confirm project-aware tools are present.
  7. Ask ChatGPT to call codencer.list_projects.
  8. For an approved one-task proof, ask it to call codencer.submit_project_task_and_wait.
  9. For an approved manifest proof, ask it to call codencer.run_project_manifest.
  10. If the response includes a blocker requiring planner decision, stop and save the blocker JSON.

Do not publish or mark live ChatGPT proof passed until client_connected, client_used_tool, and full_e2e_execution evidence is saved.