Local MCP server for AI-driven code editing and automation
boris, developed by Mjkoo, is a drop-in Model Context Protocol server that gives AI agents controlled access to local development workflows. It exposes a set of text-processing and system tools that let agents inspect, search, and edit repository files through the MCP language. The app ships as a single static Go binary suitable for laptops, containers, or CI runners. It targets developers and AI engineers building agent-based automation for code refactoring and localization workflows.
What tasks can you actually use it for?
The tool focuses on automated repository inspection and repeatable edits driven by an agent. It provides shell command execution with streaming output, file viewing with line numbers and ranged reads, project-wide regex search, and an AI-oriented string-replace utility that performs targeted edits across files. Use cases include automated localization passes and bulk code refactors, where scripted transformations are applied consistently across many files.
- Bash execution with streaming output
- Regex-based grep across repository files
- Project-wide string replacement
Is it practical to run inside development pipelines?
Deployment emphasizes portability: you build the server with go build and run a single static Go executable on desktops, Docker containers, or CI runners. The server supports STDIO for local integrations and HTTP with server-sent events for remote or containerized agents, which enables streaming responses into pipelines. This allows embedding into CI jobs or developer machines without additional runtime dependencies, provided the pipeline can run the binary and an MCP-compatible client.
What input constraints shape its effectiveness?
Effectiveness depends on correct file selection and an MCP client. Glob matching follows .gitignore rules and grep accepts regular expressions, so results hinge on accurate patterns and repository layout. The server offers an Anthropic compatibility flag to surface schemas for certain models, which affects client integration. Incorrect globs or regexes can yield incomplete or overly broad edits, so validate patterns on small samples before applying changes broadly.
Does it give you control over security and remote access?
Security is implemented through explicit scoping and authentication options. The server enforces path scoping and supports --allow-dir and --deny-dir flags to restrict filesystem access, while HTTP mode accepts bearer tokens via --token for authorized remote clients. STDIO mode keeps interactions local to the host process. These controls let operators limit agent reach and require deliberate configuration before granting edit rights, which supports auditability in team environments.
Who should adopt it, and where it falls short
boris is a practical option for development teams and AI engineers who need programmatic, auditable agent access to codebases. It requires operator familiarity with MCP clients and hosting a server binary, and automated edits should receive human review before integration. Adopt it for repeatable, scriptable repository changes and enforce narrow scopes during initial runs to reduce accidental broad modifications.
