What stays familiar

Symbol and full-text search, cross-language navigation, and code intelligence are all present. The query model is similar: ask for callers, ask for definitions, navigate across files.

What changes

The index is rebuilt on push, not in a background batch job, so there is no lag between a push and the first searchable result. There is no second cluster to run. Every query through the MCP server is attributed to a scoped token and lands in the audit log.

Feature comparison

Dimensionsrc.landSourcegraph
Code searchSymbol and text search via MCP tools and the web UI. Graph updated on push, not batch.Cross-repo text and structural search (Comby). Rich query language. Depends on batch indexing.
Indexing modelKnowledge graph rebuilt incrementally on push. Symbols, callers, and cross-language edges.Batch indexer (zoekt). Index lag between push and search availability.
Git hostingFull git host. Push, pull, SSH and HTTPS, branches, tags, web UI, issues, PRs.Not a git host. Connects to GitHub, GitLab, or Bitbucket and indexes those repos.
AI / agent accessMCP protocol server. Agents query the graph directly with scoped bearer tokens.Cody AI assistant with a proprietary context API. No standard MCP server.
Audit logAppend-only row per action. Every agent query attributed to a token and namespace.Audit logs on Enterprise. Covers admin events; not per-query agent attribution.
Permission modelNamespace tree. Agent tokens scoped per repo and action type.Mirrors upstream host permissions. No independent scoping of agent queries.
InfrastructureOne Go binary plus Postgres. No separate search cluster.Requires its own cluster alongside the existing git host. Two systems to maintain.
AuditabilityWho queried which symbol at what time is attributable to a token and namespace.Search queries tied to Sourcegraph user accounts, not the upstream git identity.
Self-hostingSingle binary. Air-gap install with offline license.Helm or Docker Compose. Requires running alongside existing git infrastructure.

Migration path

  1. 01

    Import your repos

    Paste any HTTPS or SSH git remote URL. History and branches arrive in full.

  2. 02

    Graph built on first import

    Symbols, callers, and cross-language edges index from the imported history automatically.

  3. 03

    Point agents to the MCP server

    Update your AI tooling to use the src.land MCP server URL with a scoped bearer token.

  4. 04

    Retire the Sourcegraph cluster

    Once symbol search and code intelligence are verified, you can decommission the separate search cluster.

If you are currently on GitHub or GitLab plus Sourcegraph, the migration consolidates two systems into one. Import history, verify the graph covers your key repos, then decommission the search cluster.

Try it yourself

One repo is enough to evaluate.

Import from any git remote. History and indexing arrive together.