What a cloud agent gets
Twenty-three tools over MCP: twelve from codesearch and eleven from locallm. The agent
calls them itself, whenever a question is cheaper to ask locally.
| Tool | What it does |
|---|---|
search_code |
search by meaning and by exact name at once |
get_code_chunk |
pull a match in full |
go_to_definition |
where the symbol under the cursor is declared |
find_references |
who calls it and where it is used |
find_implementations |
who implements or overrides it |
find_relationships |
the relationship graph: implementation, override, type |
analyze_impact |
direct C# impact evidence with source coordinates, hashes and snapshot |
lsp_open_document |
hand navigation the text of an unsaved file |
lsp_close_document |
give navigation back to the indexed snapshot |
ask_local |
mechanical work over files you name |
triage_log |
make sense of machine output of any length |
read_image |
read images from disk or one PDF page by page |
index_documents / search_documents |
index scanned PDFs and find OCR phrases with file versions and page references |
translate_local |
translate, structure intact |
index_status |
whether an index exists, what built it, whether the mainline branch has moved since |
index_refresh |
refresh the index after a commit |
index_unload |
free the memory a loaded index holds |
local_models_status |
what is installed, what is resident, what is missing |
local_model_preflight |
prove where a model loads: entirely in video memory under the default policy |
local_models_sync |
fetch the recommended models that are missing |
local_model_experiment_report |
statistics for one task-and-model pair |
local_model_feedback |
decide a model’s fate: promote, keep as fallback, disable |
← Two ways to use it · Contents · What the command line gets →