clojure-mcp

Executes Clojure code directly against a running nREPL server with automatic error repair capabilities. Uses a multimethod-based tool system that sends code to the REPL, captures output/errors, and applies heuristic-based fixes (e.g., missing imports, syntax corrections) before re-evaluating. This enables AI assistants to iteratively refine code within the live development environment without round-tripping through file saves.

Provides structured code editing via two complementary tools: clojure_edit for full-file transformations and clojure_edit_replace_sexp for surgical S-expression replacement. Uses tree-sitter or similar AST parsing to identify and replace specific S-expressions by pattern matching, preserving formatting and context. Integrates with file write safety checks to prevent accidental overwrites and validates syntax before persisting changes.

Implements a multimethod-based tool system where each tool registers implementations for five core multimethods: tool-name, tool-description, tool-input-schema, tool-execute, and tool-category. This architecture enables dynamic tool registration, composition, and execution without tight coupling between tools. Tools are discovered and invoked through a unified dispatch mechanism, allowing new tools to be added by implementing the multimethod interface.

Analyzes Clojure project structure by inspecting the file system, reading deps.edn/project.clj, and querying the nREPL for loaded namespaces and dependencies. Exposes project metadata including source paths, dependencies, and namespace topology through a structured inspection tool. Enables AI assistants to understand project layout and make context-aware decisions about code generation and refactoring.

Implements a configuration system that reads .clojure-mcp/config.edn files to selectively enable/disable tools, prompts, and resources at runtime. Uses a multimethod-based tool registration system where each tool is registered conditionally based on configuration predicates (tool-id-enabled?, prompt-name-enabled?, etc.). Supports directory allowlisting to restrict file system access and feature flags for bash execution and scratch pad persistence.

Executes shell commands via a bash tool that can route execution either directly to the OS shell or through nREPL's bash-over-nrepl capability (configurable via get-bash-over-nrepl). Captures stdout/stderr and exit codes, enabling AI assistants to run build tools, package managers, and system utilities. Respects directory allowlists to prevent arbitrary file system access.

Provides file read/write operations (read_file, file_write) with pattern-based search capabilities (grep, glob_files, LS). Uses ripgrep for efficient text search and respects directory allowlists to prevent unauthorized file access. Implements write safety checks to validate file paths and prevent overwrites of critical files. Supports reading files with pattern matching to extract specific sections.

Implements the core MCP server using a factory pattern where build-and-start-mcp-server coordinates startup with factory functions for tools, prompts, and resources. Uses the multimethod-based tool system to dynamically register tools at server initialization, with each tool implementing five core multimethods (tool-name, tool-description, tool-input-schema, tool-execute, etc.). Manages server lifecycle including initialization, tool registration, and shutdown.

Exposes AI prompts and resources through the MCP protocol, allowing AI clients to discover and use context-specific prompts for Clojure development. Implements prompt filtering via configuration (prompt-name-enabled?) to selectively enable prompts per project. Resources are registered similarly, enabling AI assistants to access project-specific documentation or templates.

Provides optional scratch pad functionality (controlled via get-scratch-pad-load configuration) that persists REPL state across sessions. Enables AI assistants to maintain context and intermediate results across multiple interactions. Integrates with nREPL to save and restore namespace state, allowing long-running development sessions to preserve accumulated definitions and data.

Integrates with cljfmt (Clojure code formatter) to enforce consistent code style across generated and edited code. Uses configuration (get-cljfmt) to enable/disable formatting. Applies formatting rules to code before writing to files, ensuring that AI-generated code adheres to project style guidelines. Supports custom cljfmt configuration files.