JarvisMCP
One gateway, two tools, hundreds of capabilities
Inside the project
01 / 01JarvisMCP project artwork: many capabilities routed through one gateway.
The brief
A problem worth
building around.
The standard way to give an AI agent capabilities is one tool per endpoint. That approach collapses at scale: every tool definition is spent from the context window before the agent has done anything, so a few dozen integrations can consume thousands of tokens purely describing what is available. Worse, each new server is another process to run, another credential to distribute, and another place for a secret to leak into a model's reasoning trace.
Our approach
JarvisMCP inverts the model. Instead of exposing N tools, it exposes two — one to search a ranked capability catalogue, one to execute code — and lets the agent write JavaScript against a single unified SDK. That code runs inside a workerd V8 isolate under a hard timeout, so arbitrary agent-authored programs are contained by construction. The security property that matters is the boundary: upstream credentials for every integrated system live only in trusted processes on the far side of a loopback line, and nothing the agent writes can reach across it. The isolate can ask for work to be done; it can never see the key that does it.
The experience
What it lets
people do.
The capabilities that turn the underlying engineering into a usable product.
- 01
Find the right capability
A discovery layer lets an agent look up the tools relevant to a task. It keeps a broad integration surface navigable without putting every tool definition into every interaction.
- 02
Connect actions into a workflow
Code Mode lets an agent compose calls and use their returned data in the next step. The gateway brings those integrations behind a consistent execution interface.
- 03
Keep the work accountable
Governed access and durable coordination give ongoing work a place to record ownership, progress and evidence. An agent conversation can hand off its work without becoming the only record of what happened.
Project record
What came out of it.
62 services and 558 discoverable methods reachable through exactly 2 tools
Capabilities discovered on demand instead of loading every endpoint into the model context
Arbitrary agent-written JavaScript contained in a V8 isolate under a hard timeout
No upstream credential ever crosses into the sandbox — secrets stay in trusted processes
Batching many calls into one execution turns per-call round-trips into a single request
Deployed on Docker Swarm with bearer-token auth and per-key rate limiting
Under the hood
Two tools in front of a sandboxed isolate; every upstream credential stays in trusted processes the sandbox cannot reach.
Two tools in front of a sandboxed isolate; every upstream credential stays in trusted processes the sandbox cannot reach.
Facing Similar Challenges?
Every business is different, but the problems tend to rhyme. Get in touch and tell us about yours.