The tool set is deliberately ordered around search: semantic_search and grep_search are meant to be an agent's default, with the slower stateless chat kept as a last resort because it can take far longer than pulling the right artifacts. The distinction between the two searches is the useful part — grep matches filenames and paths literally, so it finds a config by name even when nothing inside it says so, while semantic search handles the 'where is the auth code' questions.
A server fronting CodeAlive's context engine for large codebases: 11 tools spanning semantic and literal search, file reads, repository orientation, call-graph relationships and metadata analytics over your indexed repositories.
- get_data_sources lists your indexed repositories and workspaces, so the agent knows what it can search before it starts
- semantic_search is the canonical retrieval across indexed artifacts, and grep_search does exact literal or regex search inside file content plus filename and path matching — returning a file like Form.xml even when its content never mentions the name
- get_repository_ontology gives repository-level orientation, get_file_tree inspects a bounded tree for one repository, and read_file reads a repo-relative path with an optional line range
- fetch_artifacts loads the full source for search hits and reports back identifiers it could not find rather than dropping them silently
- get_artifact_relationships expands the call graph, inheritance and reference edges around one artifact, so the agent follows structure rather than guessing
- get_artifact_query_schema shows the supported ArtifactQuery entities and fields, and query_artifact_metadata runs read-only metadata analytics across selected repositories
- chat is a slower, stateless synthesized Q&A over the codebase — a fallback the docs say to reach for only when a multi-step retrieval workflow will not do
A CodeAlive API key, held in CODEALIVE_API_KEY, obtained from app.codealive.ai under MCP & API. The fastest path is the remote server at https://mcp.codealive.ai/api with no install; to run it yourself, use the ghcr.io/codealive-ai/codealive-mcp image over stdio, or self-host the HTTP transport.
One command plus a key — claude mcp add --transport http codealive https://mcp.codealive.ai/api, then supply credentials