Plans grounded in your handbook
OneMCP reads your API specification, documentation, and policies so every plan stays aligned with the operations and parameters you already maintain.
Open-source OneMCP runtime
Cache the intelligence. Serve the action.
OneMCP turns natural-language prompts into cached execution plans so agents fulfil API requests instantly—with enterprise-grade accuracy, cost control, and performance.
Plays nicely with your AI stack
Anthropic
OpenAI
Gemini
Azure OpenAI
AWS Bedrock
LangChainFireworks.ai
MCP-native
Why OneMCP
A compiled interface for every agent that touches your API.
Model Context Protocol solved connectivity. OneMCP solves the rest—accuracy, latency, and cost—by transforming prompts into cached execution plans. Agents get a natural-language surface; your systems get deterministic automation with observability, governance, and reuse built in.
Inference where it matters, cache everywhere else
Generate an execution plan a single time, then replay it indefinitely. Similar prompts reuse cached logic without touching a model.
CPU-speed automation with deterministic outputs
Deploy warm caches to keep production workloads fast, predictable, and dependable no matter the volume of requests.
From handbook to production cache
Four steps turn your API into a reusable execution engine for every agent you ship.
Import the handbook
Ingest your API specification, docs, and authentication details to establish complete operational context.
Generate execution plans
OneMCP interprets incoming prompts, builds multi-step plans, and executes them safely against your API.
Reuse cached logic
Cached plans are retrieved for similar prompts, removing redundant reasoning and stabilising latency.
Deploy with a warm cache
Export the cache alongside the runtime so production runs without inference, with predictable cost and behaviour.
What cached execution delivers
Expose every API operation through a single natural-language surface
Cache execution plans for similar prompts and replay them instantly
Pre-warm deployments so production runs without live inference
Optimize for your runtime
Switch between static and dynamic plans without changing how agents interact with your API.
Recommended for production
Static mode
Serve only prebuilt or cached execution plans. No runtime inference, fully deterministic behaviour, and effortless governance for regulated environments.
- Zero model footprint once deployed
- Versioned plans you can audit or roll back
- Stable performance even under heavy load
Perfect for exploration
Dynamic mode
Generate new plans on the fly when a prompt is unseen. Every plan is cached for reuse so experimentation turns into production-ready assets.
- Natural-language prototyping with instant feedback
- Caches grow automatically as teams iterate
- One interface across development and production
Example workflow
Prompt → plan → replay
“Create an account for Jane Doe with abc@def.com and set her up as a VIP member.” OneMCP interprets the intent, generates the correct API sequence, and stores that plan. The next time a similar request arrives, execution happens at cache speed—with full logging, policy enforcement, and deterministic output.
- Prompt interpretation aligned with your handbook
- Automatic plan storage and versioning
- Replay without inference or additional cost
Open source core
Ship the full MCP runtime, cache, and plan export pipeline under an open licence so your stack stays portable.
Governance ready
Plug in tracing, policy enforcement, and audit tooling. Enterprise extensions add observability, feedback loops, and optimization.
Built for every agent
Works with any MCP-capable agent or orchestrator, ensuring compatibility as new LLMs and frameworks arrive.
Engineered for growth
“APIs are finite and predictable. Once execution plans exist, OneMCP delivers the speed of compiled code with the flexibility of natural language.”
90%less inference spend with cached plans
<120msplan replay latency in static mode
1 toolto expose your entire API through MCP
24/7deterministic execution with full observability
Start building with OneMCP today
Read the handbook guide or clone the repo to create your first cached execution plan.
Launch the docs