Quickstart
From API key to first outline in a few minutes. Two moving parts: the binary, and a language server for your ecosystem on PATH.
1.Install the binary
Grab it from the download page (checksums there), put it on PATH, then confirm the license machinery works end to end:
context-engine license-check
2.Check your language servers
Context Engine drives the language servers you already have — it does not bundle them. Whether one is on PATH depends on your ecosystem:
| Ecosystem | Typical situation | If missing |
|---|---|---|
| Rust | Usually already on PATH via rustup. | rustup component add rust-analyzer |
| Go | gopls is usually present for Go developers. | go install golang.org/x/tools/gopls@latest |
| TypeScript | Editor-bundled servers (VS Code) are NOT on PATH as standalone binaries — install one. | npm i -g typescript-language-server typescript |
| Python | Same pattern: editor extensions don't put a server on PATH — install one. | pip install python-lsp-server |
(For Python, pyright works too if you prefer it.)
3.Set your key
Get a free key from the dashboard and export it once in your shell profile — every client snippet below reads it from the environment, so it never lives in a config file:
export CONTEXT_ENGINE_API_KEY="<your key>"
4.Register the MCP server in your client
Connect your client
The snippet reads your key from the CONTEXT_ENGINE_API_KEY environment variable — set it once in your shell profile.
# 1. Put your key in your shell profile: export CONTEXT_ENGINE_API_KEY="<your key>" # 2. Register the server: claude mcp add context-engine -- context-engine serve
5.First run: what to expect
The first outline on a large repository waits for the language server's initial indexing — rust-analyzer may take minutes on first open of a large repo. Subsequent outlines are fast; the engine keeps the server warm across calls.
6.Troubleshooting
If a first call fails silently, the usual cause is a language server missing from PATH (step 2). Context Engine reports missing or failed language servers in the semantic tool response. Install or correct the reported server, then call reload_config and retry.
To watch what the binary is doing: context-engine serve --log-file /tmp/context-engine.log writes a session log, and lsof -p <pid> shows exactly which processes and connections it holds. Still stuck? The FAQ covers the common questions.