tari-project / tari / 16425865230

62%

Build Type

push

github

Committed by web-flow

Commit Message

feat: mcp servers (#7220)

Description
---
This feature introduces a comprehensive implementation of the Model
Context Protocol (MCP), enabling secure AI agent integration with the
Tari network. It adds three new crates to the workspace:

1. `minotari_mcp_common`: A robust, shared framework for building MCP
servers. It includes a security-first architecture with local-only
binding, permission levels (read-only vs. control), rate limiting, audit
logging, input sanitization, and utilities for auto-launching and
managing dependent Tari processes.
2. `minotari_mcp_node`: A new MCP server application that exposes
read-only base node functionality to AI agents. This allows AI to query
blockchain state, network status, and mempool information safely.
3. `minotari_mcp_wallet`: A new MCP server application that provides AI
agents with access to wallet operations. It supports both read-only
queries (e.g., checking balance) and protected control operations (e.g.,
sending funds), with an optional user confirmation layer for added
security.

Additionally, this PR includes:
- A minor fix in the console wallet's gRPC server to correct a filter
condition (`.clone()` was removed from a `Copy` type).
- Removal of an unstable Rust feature (`internal_output_capture`) from
the integration tests.
- Extensive documentation in the main `README.md` and a new
`docs/mcp/TARI_MCP_IMPLEMENTATION.md` file explaining the architecture
and usage.

Motivation and Context
---
The primary motivation is to allow AI agents (like Claude, GPT-4, etc.)
to securely and effectively interact with Tari blockchain functionality.
Direct gRPC access is often too complex and risky for AI models. The
Model Context Protocol provides a standardized, secure, and
agent-friendly interface.

This implementation opens up new possibilities for building AI-powered
applications on top of Tari, such as:
- Automated network monitoring and analysis.
- AI-driven blockchain exploration and reporting.
- Controlled w... (continued)

Run Details

4600 of 15255 new or added lines in 67 files covered. (30.15%)

10 existing lines in 2 files now uncovered.

74699 of 138149 relevant lines covered (54.07%)

197244.98 hits per line