MCP Won't Solve Enterprise AI Integration (We're Missing a Layer)
3 comments
The auth problem you've hit is the exact use case for OAuth relay at the proxy layer - the proxy holds enterprise tokens and handles refresh independently, so each tool integration stays stateless and free of credential lifecycle complexity. The natural language to specific ID mapping (your channel_id example) is the genuinely harder piece, though semantic routing at the middleware layer helps: the proxy selects the right tool based on semantic matching, which reduces how often the LLM needs to reason about entity IDs directly. What's missing isn't MCP features - it's an opinionated middleware layer that decouples auth handling and semantic routing from individual tool implementations. Some open-source proxy projects are building exactly this, though enterprise connectors for Teams and Salesforce at scale are still thin.
The missing middleware layer is real... MCP handles the wire format but you still need something managing OAuth flows, token refresh, rate limiting, schema mapping per user etc... all the operational stuff. Authentication alone is brutal when you've got dozens of different OAuth implementations each with their own issues around scopes and refresh logic which is where most integrations get bogged down
Youre just missing workflows (agentic code) by trying to outsource everything to the model. MCPs are not workflows…
Slack. Google. Microsoft. Salesforce. Reddit!?
I thought: finally — a standard way for AI to integrate with enterprise tools.
So I started building an enterprise MCP gateway.
Simple use case:
30,000 employees running Copilot or Claude.
All connecting to MCP tools.
Step 1: build a gateway.
Step 2: connect directory.
Step 3: assign MCP tools to users.
So far so good.
Then reality started stacking up.
Problem #1
You can’t let 30,000 employees authenticate directly to every MCP endpoint. So the gateway uses admin credentials.
Congrats.
Now your AI system technically has access to every Teams message in the company.
Problem #2
LLMs reason in natural language.
MCP tools expose REST wrappers.
Nancy asks:
“Summarize the marketing channel from yesterday.”
The tool expects:
get_messages(channel_id=847239)
So now you’re dynamically mapping IDs to names and rebuilding tool schemas per user.
Problem #3
OAuth tokens expire.
Now your gateway is refreshing tokens, retrying calls, translating requests, rebuilding responses, and basically turning into a giant middleware monster.
At this point I realized something:
MCP isn’t the problem, Nancy is not the problem either.
MCP It’s actually great.
But the industry is trying to use it to solve the wrong layer of the problem.
Trying to wire enterprise AI together through direct MCP tool connections is not architecture.
It’s integration chaos.
What we’re missing isn’t more connectors.
What we’re missing is ... well thats what I"m working on now, it involves abstract agent routing - like Layer 3.5 for AI.
Until then - I really care about Nancy and all the poor bastards working in large companies that will figure this out too but can't walk away because they need that two week pay.
Sense of humor but I"m making a point MCP = Missing Core Parts trying to use it on a enterprise level for AI Integration in a walled garden its just not going to work.