Local context for coding agents
Let your coding agent spend tokens writing code, not finding it.
Measured in our three-turn coding-agent test.
Avg. across our coding-agent test
In our three-turn experiment
Index, retrieval, and source code
Apache 2.0 on GitHub
Before your agent can write code, it spends context searching, opening files, and following connections.
The agent searches broadly before it knows where to work.
One useful file leads to the related code the change depends on.
session.ts
createSession()
middleware.ts
requireAuth()
user.ts
type User
token.ts
verify(jwt)
store.ts
readSession()
auth.ts
no references
Scubiee gives coding agents focused repository context before they start writing.
Search by meaning and surface the files that matter to the task.
Follow relationships between files without opening the whole repository.
Return focused context instead of large, token-heavy file dumps.
Refresh changed files incrementally as your codebase evolves.
Keep your repository index and source code on your machine.
Connect to the coding tools you already use through MCP.
MCP tools
Scubiee exposes a phase surface over MCP — semantic map first, focused reads second, grep when you need literals. Recommended flow: gate → map → focus → edit → sync if needed.
gate / status
Is this repo ready?
One check at session start — managed, warming, or needs init/connect.
map
Where is X handled?
Ranked overview of paths and symbols — no full file bodies.
focus
Show me that handler
Deep-dive spans, neighbors, and call sites from a map hit.
grep
Find every API_KEY
Exact literals and regex across indexed files when you know the string.
glob
List all *test*.py
Path patterns over the index — faster than walking the tree blindly.
workspace
What did we already read?
Session memory — pins and heatmap so agents do not re-explore.
Recommended agent flow
status(root=workspace) → map(query) → focus(target) → edit → scubiee sync .
Give your agent a shorter path from a question to the code it needs.
7 → 4 steps
Fewer retrieval steps
60–70%
Fewer context tokens used
More room
Left for actual coding
Scubiee builds and uses its repository index locally, so your source code does not need to leave your computer.
Keep the repository, index, and retrieval loop inside your local environment.
Local repository index
Connect Scubiee to compatible coding agents through the Model Context Protocol.
Cursor
via MCP server
Claude Code
via MCP server
Codex
via MCP server
OpenCode
via MCP server
Any MCP tool
if it supports MCP
Compatibility is provided through MCP and does not imply endorsement by, or partnership with, any agent provider.
A focused retrieval loop gives your coding agent the context it needs.
Build a local repository index — Scubiee scans your codebase and stores embeddings and relationships on your machine.
Query the index by meaning. Scubiee surfaces the files and functions most relevant to the task, not the whole codebase.
Return focused, ranked context to the agent. Less noise, fewer tokens, and more room to actually write code.
Incremental refresh — changed files update in under a second.
Your index stays current without re-indexing the entire codebase.
A measured experiment shows how focused retrieval can leave more of the context window available for coding.
Three-turn experiment
60–70% fewer context tokens
In our three-turn coding-agent experiment, Scubiee used about 60–70% fewer context tokens than the comparison workflow. Results vary by repository and task.
Ready when you are
Takes about 2 minutes to set up end-to-end.
