LSP (Language Server)
Use LSP when you know a symbol's name and want to find its definition, callers, or type.
For example: "Where is verifyJWT defined? What calls it? What is its type?"
Fit: symbol-scope navigation, refactors, type queries. Skip when the change is a one-line literal edit or when your question is about history rather than the working tree.
Solo Dev — Serena
Serena connects your agent to a language server through MCP without requiring a separate index build.
It uses the MIT license and runs through uvx.
claude mcp add serena -s project -- \
uvx --from git+https://github.com/oraios/serena serena start-mcp-server --project $(pwd)
Serena uses the project’s files to determine which language servers to run and supports more than 40 languages out of the box.
Add these instructions to AGENTS.md:
When you need a symbol's definition or its callers, use Serena's
`find_symbol` or `find_referencing_symbols` MCP tools instead of
searching files by name. Serena answers by the language's own scope rules.
For symbol-body rewrites, prefer Serena's replace/insert tools over diff
patches — they are atomic across files when the language supports it.
On a large repository, the first startup takes 30 seconds to a minute while the language server builds its index. Subsequent queries take milliseconds.
Org — SCIP
The indexer runs in CI and writes a Protobuf file associated with a specific commit. Tools that read this file can query the index offline.
- name: Build SCIP index
run: scip-<language> index --output index.scip
- name: Upload
run: aws s3 cp index.scip s3://<bucket>/scip/${{ github.sha }}.scip
Available indexers include scip-typescript, scip-java, scip-python, and scip-clang.
For a shared setup, run the indexing job on every push to main and use an MCP wrapper to
fetch the artifact.
Dive deeper
- Serena (Solo Dev):
research/serena/— scored capability matrix, per-cell reasoning · pinned lab - SCIP wrapper (Org):
research/scip/· pinned lab