Today's MCP marketplaces are walled gardens with different schemas, different trust models, and zero portability. Three standards tracks are working to change that. Here is what each is doing, where they overlap, and what to bet on.
Why standards now
Three converging needs:
- Publisher portability — list once, discoverable everywhere.
- Consumer trust — verify a server's identity regardless of marketplace.
- Tooling interop — clients should not hard-code per-marketplace logic.
Without standards, the next two years look like the Linux distro fragmentation of the early 2000s.
The three tracks
1. IETF mcp-discover (draft)
A protocol-level extension. A standard discovery endpoint, a standard descriptor format, signed advertisements. See discovery protocol.
- Scope: how an agent host finds MCP servers across networks.
- Status (mid-2026): active draft; first interop event scheduled.
- Likely outcome: RFC by 2027.
2. W3C Agent Capabilities Community Group
Higher-level than discovery. Defines a vocabulary for what agents can do, how they describe themselves, and how a host can ask the right one to act.
- Scope: semantic descriptions of agent capabilities.
- Status: community group; pre-WG.
- Likely outcome: a Note (recommendation) by 2027, full WG track later.
3. Community Registry Interop Spec
Registry-to-registry sync. Lets a server published on Loadout appear on Smithery without re-submission.
- Scope: registry APIs and metadata mapping.
- Status: GitHub-based working group; reference implementation in flight.
- Likely outcome: published spec late 2026.
What they cover, side by side
| Concern | IETF mcp-discover | W3C Agent Caps | Registry Interop |
|---|---|---|---|
| Find the server | Yes | No | Indirectly |
| Describe what it does | Basic | Yes | Yes |
| Verify identity | Yes (signing) | No | Mirror |
| Cross-network | Yes | N/A | Yes |
| Reputation | No | No | Optional |
The three tracks are complementary, not competitive.
What each unlocks
- mcp-discover: browser-style "find a tool for this task" without static configs.
- W3C agent caps: an agent can ask another agent "are you the right one for this?".
- Registry interop: publishers list once, consumers see them everywhere.
Together they unwind the lock-in of today's marketplaces.
Where today's marketplaces sit
Three positions in the landscape:
- Anthropic Hosted Tools — Anthropic is participating in IETF; their format aligns with mcp-discover early.
- Smithery — building registry interop; sees portability as competitive advantage.
- mcp.so — community-aligned; likely interop-first.
Walled-garden registries that refuse to participate will lose ground starting in late 2026.
What publishers should do today
Three forward-compatible moves:
- Use the draft mcp-discover descriptor format for any registry submission. It is forward-compatible.
- Sign your packages. Every standards track ends up requiring signing.
- Publish a
.well-known/mcpendpoint if you host your own server.
These are days of work and avoid weeks of migration when standards land.
What consumers should do today
- Build with abstraction layers — do not couple your client code to one registry's API.
- Track the working groups — major changes are signposted months ahead.
- Demand interop — when your vendor's marketplace does not federate, ask why.
Risks
Two failure modes for the standards process:
- A vendor forks — common in early standards, slows adoption by 12+ months.
- Premature lock-in — early implementations bake in choices the standard later changes.
Mitigate by keeping abstractions thin and tracking each draft.
Common mistakes
- Hard-coding to one registry — locks you in.
- Ignoring signing — all roads lead to it.
- Building a "marketplace" without interop — does not survive 2027.
- Waiting for the standard to land — the standards are stable enough to build on now.
Where this is heading
Three predictions for 2027:
- mcp-discover ratifies; major hosts implement.
- Registry interop produces the first real cross-marketplace experiences.
- W3C work matures; agent-to-agent capability negotiation becomes routine.
Build to the drafts; you will be early but compatible.