Context+ MCP Server: Setup for Claude, Cursor and Other Apps

• Finds code by what it does, not by exact words

  Matches your question to files and symbols by meaning, so "user authentication" can find login and session code that never uses those words. It returns the matching definition lines. It builds and saves a search index in the project, and sends code snippets to your embedding provider.

  Try: “Find the code that handles user authentication.”

  Write · semantic_code_search

• Finds functions and classes by intent, with where they are used

  Looks up functions, methods, classes and variables by what you describe, then lists their definitions and the places that call them, with line numbers. It uses the same saved embedding index as code search.

  Try: “Where is the retry logic for failed payments defined and called?”

  Write · semantic_identifier_search

• Shows the project's folders, files and symbols as a tree

  Draws the project as a tree with each file's short header and the functions, classes and enums inside it. Large projects are trimmed automatically to fit a token limit, and you can limit depth or point it at one folder.

  Try: “Give me an outline of the src folder, two levels deep.”

  Files · get_context_tree

• Shows a file's function signatures without the bodies

  Lists the functions, methods and types in one file with their parameters, return types and line ranges. It lets the assistant learn how to use a file without loading all of it.

  Try: “Show the public API of src/core/parser.ts.”

  Files · get_file_skeleton

• Groups the project's files into labeled topic clusters

  Clusters files that are about similar things and gives each cluster a name using a chat model. It needs an embedding model and a chat model, and it saves embeddings in the project.

  Try: “Group this repo's files by feature and name each group.”

  Write · semantic_navigate

• Lists everywhere a function, class or variable is used

  Finds every file and line that imports or uses a symbol you name. Use it before deleting or changing code, and it warns when a symbol is barely used.

  Try: “What breaks if I rename parseConfig?”

  Files · get_blast_radius

• Runs the project's own linter or compiler and reports problems

  Runs a checker that matches the file type: tsc for TypeScript, eslint for JavaScript, py_compile for Python, cargo check for Rust, go vet for Go. These run on your machine inside the project folder with a 30 second limit.

  Try: “Run the type checker on this project and list the errors.”

  Runs code · run_static_analysis

• Saves a file after checking it against the server's code rules

  Writes a full file to the project, but only if it passes the server's rules, such as a two-line header, no inline comments, a nesting limit and a file length limit. It saves a restore point of the old file first.

  Try: “Save this new version of src/api/client.ts.”

  Write · propose_commit

• Lists the saved copies made before the assistant changed files

  Shows every restore point made before a save, so you can pick one to go back to. Reads a small file in the project's .mcp_data folder.

  Try: “Show the restore points from today.”

  Files · list_restore_points

• Puts files back to how they were before a specific change

  Rewrites the files from a restore point over the current ones, so changes made after that point are lost. Git history is not touched.

  Try: “Undo the last change to src/api/client.ts.”

  Write · undo_change

• Browses feature maps written as Markdown notes with links to code

  Reads hub notes, which are Markdown files with [[wikilinks]] to code files. It can list the hubs, show one with outlines of the files it links, or find files that no hub mentions.

  Try: “List the feature hubs and any files that are not linked to one.”

  Files · get_feature_hub

• Saves or updates a note, concept or symbol in a memory graph

  Stores a node for a concept, file, symbol or note in the project's memory graph, or updates it if the same label exists. Its text is turned into an embedding using your provider.

  Try: “Remember that the payments module depends on the billing API.”

  Write · upsert_memory_node

• Links two memory notes with a typed connection

  Connects two memory nodes with a relation such as depends_on or implements. The link weight fades over time unless it is refreshed.

  Try: “Link the payments note to the billing API note as depends_on.”

  Write · create_relation

• Searches saved memory by meaning and follows its links

  Finds the notes closest in meaning to your question, then also returns notes linked to them one or two steps away, with relevance scores.

  Try: “What do we remember about the billing API?”

  Files · search_memory_graph

• Cleans old, faded links and unused notes out of memory

  Deletes memory links whose weight has faded below a threshold, and notes that have no links, are rarely used and have not been touched for over 7 days. It only touches the memory graph, not your source files.

  Try: “Clean up stale links in the memory graph.”

  Delete · prune_stale_links

• Adds many notes at once and links the similar ones

  Saves a batch of notes and automatically links any two that are similar enough, including links to notes already stored. It computes embeddings with your provider.

  Try: “Import these ten notes about our auth flow into memory.”

  Write · add_interlinked_context

• Starts from one memory note and explores what it links to

  Takes a memory node and returns all notes reachable within a depth limit, ranked by link strength and distance.

  Try: “Show everything connected to the payments note within two steps.”

  Files · retrieve_with_traversal

Watch out for

• propose_commit overwrites any file the path points to

  It writes whatever path the assistant gives, and the 1.0.8 build does not check that the path stays inside the project folder, so a path such as ../../.zshrc is written as given. The header and comment rules only reject some content. undo_change also rewrites files from a restore point. Work in a git branch and review every change. github.com

• run_static_analysis runs commands in your project through a shell

  It runs tsc, eslint, py_compile, cargo check or go vet with exec, joining the file path into a shell command line without quoting. cargo check and go vet run project build code, and a crafted path can add shell commands. It has no read-only mode. Use it only on projects you trust. github.com

• A cloud provider receives your code snippets

  If you set the embedding provider to OpenAI, Gemini or another API, file contents and symbol text are sent there to be embedded. The default Ollama setup keeps this local unless you add an Ollama Cloud key. github.com

• It saves data inside your project

  A .mcp_data folder holds embedding caches, restore points and the memory graph. Add it to .gitignore so it is not committed. github.com

• prune_stale_links deletes memory data

  It removes faded links and unused notes from the memory graph. It does not touch your source files. github.com

Security advisories