Context Engine: code search and memory for an agent
Context Engine
A skill that teaches an agent to query the Context Engine code index and symbol graph before reading files directly
- 400 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The background bridge process indexes the workspace and tracks changes, and an outside service processes the data
- The sign-in data sits in ~/.ctxce/auth.json and is shared by the bridge process and the MCP server
- The memory tool keeps knowledge between sessions and returns it to the agent as context
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/context-engineDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The background bridge process indexes the workspace and tracks changes, and an outside service processes the data
- The sign-in data sits in ~/.ctxce/auth.json and is shared by the bridge process and the MCP server
- The memory tool keeps knowledge between sessions and returns it to the agent as context
This is third-party code. Review the repository files before installing.
What it does
Context Engine is a service with semantic code search, memory and symbol analysis for AI assistants. The repository holds skills and rules that explain to an agent how to use more than thirty of the service's tools over MCP. The core rule from SKILL.md: when the context-engine MCP server is connected and the workspace is indexed, the agent calls repo_search first for any question about the code, and only then ls, find, Grep or file reads. For how and why questions that span several files there is context_answer with citations. symbol_graph finds definitions, callers, callees, importers and references. graph_query builds impact chains before a change and cross_repo_search looks in other indexed repositories. The skills also teach the batch queries batch_search, batch_symbol_graph and batch_graph_query to save tokens, and keeping knowledge across sessions with memory_store and memory_find. The agent also learns to search structural patterns such as retry loops and error handling, and to search Git history. A separate block for CLAUDE.md or AGENTS.md carries the rules to subagents.
Who it is for. For developers on large codebases or with several repositories who want Claude Code, Cursor or Codex to find code through an index and not page through folders.
Good fit when
- The codebase is large and the agent spends tokens combing through files with find and Grep
- Before a rename or signature change you need every use of a symbol in one query
- A how-does-it-work question spans several files and you want an answer with citations and line numbers
- Code is spread across several repositories and you are looking for the boundary between them
- You want to keep agreements and findings between agent sessions
Not a fit when
- The code must not go to an outside service: the workspace is indexed by the Context Engine service
- The project is small and plain file reading is enough for the agent
- You cannot create an account and a key: with no MCP server connected the skill searches nothing
- You need fully local work with no internet and no background processes
Example request
Find every place that calls the order creation function and show what would break if its signature changed.Limitations
The skill searches nothing by itself. It works only with the context-engine MCP server connected, and the server needs an account at dev.context-engine.ai and an API key. A background bridge process or a VS Code extension indexes and tracks the workspace, and where the indexed data goes and how long it is kept is not described in the repository.
The repository holds the client side of a hosted service. Check terms and limits at context-engine.ai. The repository ships rules for different assistants: the .cursorrules file for Cursor, GEMINI.md for Gemini and the .augment folder for Augment. For Windsurf the instructions suggest copying the folder with the Codex skills. Access to the service from Russia was not checked for this card.
How to disable. Remove the plugin from Claude Code through the /plugin menu, or the skill folder from .claude/skills. Stop the background process with ctxce stop and remove the context-engine server from the client configuration. Revoke the API key in the service dashboard if needed.
FAQ
How do I install the Context Engine skill in Claude Code?
Run /plugin marketplace add Context-Engine-AI/Context-Engine, then /plugin install context-engine. The skill is pulled from GitHub and loads tool guidance into the session. By hand you can copy the skills/context-engine folder into your project's .claude.
Do I need an account for the skill to work?
Yes. The skill only explains to the agent how to use the tools, and the tools themselves come from the MCP server tied to your account. An API key is needed for the connection, and the terms and limits are on context-engine.ai.
How do I connect a project without VS Code?
Install npm install -g @context-engine-bridge/context-engine-mcp-bridge, then run ctxce connect <key> --workspace /path/to/project --daemon. The process indexes the folder and watches for changes every 30 seconds; the --interval flag changes that. Check status with ctxce status and stop with ctxce stop.
How do I connect Context Engine to Cursor and Codex?
For Cursor copy the .cursorrules file into the project root and the rules are picked up automatically. For Codex ask it to install the context-engine skill from the repository page, or copy the .codex/skills/context-engine folder into ~/.codex/skills.
Which tools does the skill tell the agent to call first?
In SKILL.md it is repo_search for search by meaning, error text or name, context_answer for how and why questions across several files, and symbol_graph for callers and references. Before an edit it requires fetching all affected symbols in one call.
Discussion
No questions or reviews yet.
Related
SkillsA local MCP server that indexes your code so the agent pulls the functions it needs instead of rereading whole files
Fast semantic code search for agents: returns the exact snippets instead of reading whole files and spends far fewer tokens
14 skills that find the gaps in a task before the code: interviews, plans, assumption tests and proof that a bug fix works
A Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
An MCP server that turns a large codebase into a semantic feature graph for precise navigation and meaning-based search
Tech Leads Club Agent Skills
Agent Skills
A vetted skill registry with a CLI that installs into twenty-plus agents and an MCP server for on-demand skill lookup
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Context Engine: code search and memory for an agent (this entry) | yes | 400 | ||
| Claude Context | yes | API key | 12,658 | an MCP semantic search server on a vector database, with no skill or memory |
| CodeGraph | no | not required | 75,918 | a local code graph next to the project, with no external service |
README in short
The repository holds the skill files, rules for different assistants and a Node.js bridge. The license is MIT.
README badge
Original SKILL.md
--- name: context-engine description: Search and understand the indexed codebase with Context Engine (hybrid semantic + lexical search, reranked, with a symbol graph). Use for any question about this codebase - where something is implemented, how it works, callers/definitions, every place that does X - before ls/find/Grep/Read. --- # Context Engine When the `context-engine` MCP server is connected, this workspace is indexed: for any question about the codebase, call `repo_search` FIRST, before ls/find/Grep/Read, and again whenever you move to a new area; for how/why questions that span several files use `context_answer`. Results are verified source with exact line ranges: treat them as already read. Use Grep only to confirm an exact literal in a file you already know, or to count. For questions asking for every/all sites, run one repo_search per distinct pattern (queries=[...]) and follow the refs. If the tools are listed as deferred, load them with ToolSearch first. | Need | Tool | |---|---| | Find code by behavior, concept, error text or identifier | `repo_search` (`queries=[...]` for several at once; `under`, `language`, `path_glob` to narrow) | | Explain how/why across several files, with citations | `context_answer` | | Definition, callers, callees, importers, references of a symbol | `symbol_graph` (several names: `A\|B\|C`) | | Multi-hop impact before a change | `graph_query` | | Code that may live in another indexed repo | `cross_repo_search` | Before editing, fetch every involved symbol in one call; before a rename or signature change, get every site with `symbol_graph` `query_type=references`. Search again only after learning something new.