Every MCP server is a new door
Tools get exposed per team, per sprint, outside the platform. Nobody holds the list.
GRAPHQL HIVEIN YOUR CLOUD
POLICY + SCHEMA CHECKS
THE GOVERNED
SupergraphSchema-declared skills over MCP.
Subgraph
Subgraph
Subgraph
Built on GraphQL Hive and everything affiliated with it
Open source, MIT · GraphQL Foundation member · Gold sponsor, GraphQLConf 2026 · SOC 2 at company level · Core libraries at 20M+ weekly downloads
Proof
GraphQL Hive - trusted by enterprise leaders
Published case studies and customer stories on the-guild.dev.
What changed
Three years ago your graph had two consumers.
Web and mobile. Both shipped through code review, both rate-limited, both accountable to a human who could be asked what they did.
You now have a third consumer that writes its own queries, at machine speed, in shapes nobody designed for - and the fastest way to give it access is also the least governed one.
Agents don't respect API contracts
They compose calls you never specified, at volumes you never sized, against fields you never meant to expose.
“What did the agent read last Tuesday?”
If the answer needs an engineer and two days, it isn't an answer.
The position
Connecting agents is the easy part.
Exposing a tool to an agent takes an afternoon. Everything after it is the work: which fields an agent may read, which mutations it may never call, which team's budget its tokens land on, and what you put in front of an auditor.
That machinery already exists for your human consumers - schema registry, schema checks in CI, policy, usage analytics. In Hive it extends to agents without a second platform: skills are declared in the schema, exposed over MCP, and governed by the same checks. No new platform, no replatforming.
Our job is to make that real in your environment, in weeks, with the people who wrote the code in the room.
How agent governance worksWhat we do
Open-source federation - your graph, your cloud
Registry first or gateway first, subgraph by subgraph, incremental. Your production traffic does not move until you move it. Scope agreed before anything moves, with published exit criteria.
DetailsAgent readiness - the governance layer
Schema-declared skills over MCP. Field- and operation-level policy per agent. Per-agent usage, latency and cost. One audit trail covering humans, apps and agents.
DetailsPlatform engagement - the supergraph, run properly
Architecture reviews, version and gateway upgrades, performance work, and an escalation path that ends at the people who maintain the open-source project.
DetailsThe difference
You talk to the person who changes the code.
The Guild has no split between core engineering and support: the engineer on the call is the engineer who edits the router. On our engagement with you, a Guild engineer sits in the architecture sessions and the migration reviews - not a solutions consultant reading from a runbook, and not a ticket.
Ownership
What you own when we're finished
Your software
Open source, MIT. No open-core split, no gated enterprise features, no SSO tax.
Your infrastructure
The gateway runs in your cloud. Cloud, on-prem or a laptop.
Your license
With The Guild.
Your exit
Stop working with us and nothing in your stack changes.
We think that's the only honest way to sell infrastructure. It's also why there's no lock-in clause to read.
The 90 days
| Weeks | Stage | What moves |
|---|---|---|
| 1–2 | Graph audit | Inventory of subgraphs, consumers, breaking-change risk, agent surface. Nothing changes in production. |
| 3–4 | Registry and checks | Hive schema registry live, checks wired into CI. Still no traffic moved. |
| 5–10 | Gateway migration | Subgraph by subgraph, behind your flags, with rollback at every step. |
| 11–12 | Agents and handover | MCP skills declared and governed. Your team certified to run it without us. |
The last two weeks exist so you don't need us in month four. That's deliberate.
FAQ
Can we adopt Hive incrementally, or is it a big bang?
Incrementally. The registry and the gateway are separable concerns, so most teams move one first and run both side by side. Subgraphs themselves generally need no rewrite - they are spec-compliant either way.
Do we have to rewrite our subgraphs?
In the normal case, no. The work sits at the gateway and registry layer.
Is Hive actually fully open source, or open-core?
Fully open, MIT. There is no paid-only feature set, no gated SSO, and the gateway runs in your own infrastructure.
Can it run fully on-premise, in a regulated environment?
Yes - cloud, on-prem or air-gapped. The Guild holds SOC 2 at company level.
What is MCP, and why should it live in the graph?
MCP is how an agent discovers and calls tools. Declaring those tools in your GraphQL schema means the access rules, the change checks and the usage data you already have apply to agents automatically, instead of being rebuilt per team in a side-channel nobody governs.
Bring us your graph.
One call, an engineer on it, and a straight answer about whether this is worth doing for you.