MCP guide
MCP, tools, and project brains
Model Context Protocol is a common way for AI clients to connect with external tools and context. Amistio's current product model is local-runner first: it coordinates approved work while your machine owns repository access, tool setup, and execution.
What MCP is
MCP stands for Model Context Protocol. It gives AI applications a standard pattern for discovering capabilities, reading context, and calling tools through a server boundary. Instead of every AI client inventing a different integration shape, MCP gives teams a shared vocabulary for tools, resources, prompts, and permissions.
MCP is a protocol for connecting an AI assistant to useful context and actions. It does not remove the need for authorization, review, logging, or careful execution boundaries.
Why software teams care
Tool discovery
AI clients can learn which operations are available instead of relying only on pasted instructions or one-off scripts.
Context access
Relevant documents, records, or project facts can be exposed through a controlled server rather than copied manually.
Consistent boundaries
The same integration can define what the model may read, what it may call, and what still needs human approval.
Portable integrations
Teams can connect tools across MCP-aware clients without rebuilding every integration from scratch.
How it relates to Amistio
Amistio is not presented here as a first-class MCP runtime. Today, the important boundary is the local runner: the web app manages intent, project-brain records, review state, and queue status, while the runner works from the paired checkout with the tools installed on your machine.
- The project brain gives agents structured context before implementation starts.
- The local runner provides the execution harness around approved work.
- Local AI tools and custom commands remain machine-owned instead of cloud-owned.
- A future MCP integration would still need the same review, permission, and local-execution rules.
Where MCP can fit later
If Amistio adds first-class MCP support in the future, the safe shape would likely keep MCP behind explicit runner or workspace boundaries: approved project-brain context as resources, narrowly scoped operations as tools, and human approval before source-changing execution. That future work needs a separate decision and implementation plan because protocol support touches trust, auth, configuration, and runner adapters.