DevForge-MCP

List every project under the configured workspace root. Use this when the user wants to know what repos are available or switch projects.

Switch the active project by name (relative to the workspace root) or absolute path. Re-indexes and reattaches the file watcher — no restart needed. Use this when the user says 'work on <other project>' or 'switch to <repo>'.

Return the active project's name, path, indexed file count, detected architecture, and last scan time.

Re-scan the active project and rebuild the file index + architecture profile. Use this after large external changes (e.g. a git pull).

List indexed files with metadata (path, size, extension, last modified).

Read a single source file by relative path. Reads directly from disk — never truncated, always returns the full file content.

Use this when the user asks a coding question about the project, BEFORE answering. Returns the entire indexed codebase plus architecture summary so suggestions respect existing conventions.

Returns the detected architecture profile: language, framework, primary pattern, patterns in use, tech stack, conventions, namespace, layer map. Call alongside get_full_codebase.

Full-text search across indexed files. Use this to find relevant classes/functions/usages before generating code.

Map of every file → its imports/usings and namespace. Use before modifying or creating files to understand coupling.

Current build state: last error, last changed file, index stats, architecture summary.

Create or overwrite one file. Always call get_full_codebase + get_architecture first so the code matches the project's conventions.

Write several files in one operation — ideal for refactors spanning multiple layers/files at once.

Auto-detects the right build tool (dotnet/npm/pnpm/yarn/python/maven/gradle/go/cargo) for the active project and runs it. Returns pass/fail and full output.

Store a user/assistant turn in rolling conversation memory.

Retrieve the last 10 conversation turns for context continuity.

Find every occurrence of an identifier (variable, function, class — any symbol) across the codebase.

Find where a class/struct/interface/type is declared.

Find where a function/method is declared (C#, Java, JS/TS, Python, Go signatures supported).

Find where an interface/type-interface is declared.

Find every place a symbol is referenced (broader than find_symbol — used for 'what calls this' questions).

Heuristically flag files that no other file appears to reference — candidates for deletion. Always verify before deleting.

Detect blocks of near-identical code repeated across files — candidates for extraction into a shared function.

Surgically replace one exact, unique snippet of text in a file. The safest edit — preserves all surrounding formatting. Fails loudly if the snippet isn't unique.

Replace an entire method/function body by name, keeping the rest of the file untouched.

Insert a new method into an existing class (start or end of its body).

Append a new class/type to a file (inside the last namespace block for C#, end-of-file otherwise).

Rename a symbol everywhere it's used across the indexed codebase (word-boundary match, comment/string-safe) and write every changed file back to disk. Use scope='file' with rel_path to rename within a single file only (safer). scope='project' (default) renames across all files — always confirm with …

Remove a method/function (and its immediately preceding doc-comment/decorator/attribute lines) from a file.

Short git status + current branch for the active project.

Git diff for the working tree, optionally scoped to one file.

Recent commit history.

List files git sees as untracked.

Checkout an existing branch.

Create and checkout a new branch.

Stage and commit changes with a message.

Record a durable project-level note (an architecture decision, a coding convention, a known issue, or a free-form note) that persists across conversations, not just this chat.

Retrieve all stored project-level notes: decisions, conventions, known issues, notes.

Heuristic review of one file: long methods, TODOs, magic numbers, deep nesting.

Project-wide stats: file counts by extension, largest files, total TODO/FIXME count.

Aggregate code-smell scan across the whole project (long methods, duplication, deep nesting, TODOs), ranked by severity.

Architecture-aware suggestions based on the detected stack (e.g. missing tests, missing DI, missing validation).

Consolidated, prioritized refactoring suggestions combining code smells + architecture review.

Parse a raw compiler/build error into structured {file, line, code, message} entries (C#, TS, Java, Go, Python) so it's easy to explain in plain language. Defaults to the last build error if none is provided.