A custom MCP connector lets ChatGPT discover and invoke tools exposed by a remote MCP server. Builders need a reachable server, clear tool schemas, appropriate authentication, predictable responses and a safety model for write actions. Workspace administrators control availability in supported ChatGPT plans.
Why this matters now
MCP moves a custom integration from a one-off chat feature toward a reusable protocol surface. OpenAI documents full MCP support in ChatGPT developer mode for eligible Business, Enterprise and Edu workspaces, with administrators able to test and publish connectors. OpenAI also states the responsibility boundary: organizations remain responsible for the safety and appropriateness of servers they connect.
That is a distribution opportunity. A well-designed remote server can present the same governed capability to several compatible hosts. It is not permission to expose an internal API unchanged.
Production requirements
Names tell the model when to call a tool—and when not to.
Every request resolves to the correct user, workspace and operation.
Results carry the evidence needed for the next reasoning step.
Read and write behavior is separated; consequences require confirmation.
Use the tool-design guide before asking users to connect a server.
A sensible test path
- Begin read-only with one narrow user job.
- Test whether realistic prompts select the correct tool.
- Test authorization and cross-workspace denial.
- Inspect output size, source identity and empty results.
- Add writes only after approvals, idempotency and auditing exist.
- Run adversarial tests against connected documents.
The Apps SDK can add a ChatGPT-facing interface, but the durable investment is still the server’s data, permission and tool contract.
The SPYN3 connection
SPYN3 is a hosted MCP endpoint in front of a workspace's connected systems, usable from ChatGPT and any other remote-capable MCP client. Its current beta exposes search, query, source listing and document retrieval. See the hosted MCP quickstart for exact methods and boundaries.
Built for the next step
Give an agent your systems, not unrestricted access
Connect one SPYN3 endpoint and let it reach your workspace’s sources through read-scoped, foldered tools.
Explore the MCP endpointSPYN3 is a consolidated MCP layer: one hosted endpoint in front of every system you connect — MCP servers, Postgres, REST APIs, Google Drive — with tools foldered by source and read/write permissions per key. Switching agents means moving one URL. Free plan available; current capabilities are documented on the product facts page.
Sources
- OpenAI: developer mode and MCP apps in ChatGPT — current connector support, administration and safety responsibilities. Verified 13 Sep 2026
- OpenAI: build with the Apps SDK — MCP-based app development and testing. Verified 13 Sep 2026
- Anthropic: prompt injection defenses — indirect prompt-injection risk and layered defenses. Verified 13 Sep 2026