osfeed.dev
Use with your agentConnect an agent
Projects/zilliztech / claude-context

Claude Context

Semantic code search for MCP clients and VS Code

by@zilliztech
Get it on GitHub
Unclaimed

Written by @iluxav from the project's code and releases, not by its maintainers. Last updated , so it may lag behind newer releases.

Are you @zilliztech?Claim it, and your coding agent keeps it current with each release
Updated 13 things you can do 6 changes to know before upgrading
Follow GitHub

Find relevant code using natural-language queries in your coding agent or VS Code.

The VS Code Semantic Code Search panel shows a query field, Search and Index Current Codebase buttons, and a settings control.

What you can do with Claude Context

Find relevant code snippets with natural-language queries through an MCP-compatible coding client.

Present at v0.1.1
Good to know

Requires an indexed codebase, an embedding provider, Milvus or Zilliz Cloud, and Node.js >=20.

Index a codebase in the background, refresh changed files automatically, or clear its search index.

Present at v0.1.1
Good to know

Requires access to the codebase directory and a configured vector database.

Check whether a codebase is indexed and see progress while indexing is running.

Present at v0.1.1
Good to know

Progress is phase-based. Recovered file counts may reflect chunk counts until a full reindex; status refresh depends on database availability.

Choose additional file extensions and exclude files or directories from code indexing.

Present at v0.1.1
Good to know

Default exclusions apply. Dotfiles and dot-directories are always skipped; custom extensions and ignore patterns can be configured globally or per indexing request.

Generate search embeddings with OpenAI, VoyageAI, Gemini, or a local Ollama model.

Present at v0.1.1
Good to know

Hosted providers need API credentials; Ollama needs a running server and an installed model; vector storage is configured separately.

Search and navigate code using the Semantic Code Search extension in VS Code.

Present at v0.1.1
Good to know

Requires the extension and configured embedding and vector database services.

Generate code embeddings through a custom Gemini API endpoint.

Since v0.1.6
Good to know

Requires Gemini provider selection, GEMINI_API_KEY, and GEMINI_BASE_URL.

Show all 13

Include Dart source files in code search without configuring a custom extension.

Since v0.1.11
Good to know

Dart files use fallback text chunking rather than a dedicated Dart syntax parser.

Generate code-search embeddings using an OpenRouter model.

Since v0.1.11
Good to know

Set EMBEDDING_PROVIDER=OpenRouter and OPENROUTER_API_KEY; EMBEDDING_MODEL defaults to openai/text-embedding-3-small.

Give each codebase's Milvus or Zilliz collection a readable name prefix.

Since v0.1.11
Good to know

CODE_CHUNKS_COLLECTION_NAME_OVERRIDE is sanitized; a path hash remains appended. Setting or unsetting it selects a different collection name.

Refresh indexed code on demand from editor hooks and choose or disable automatic polling.

Since commit 6fc318b
Good to know

Trigger watching is enabled by default, uses a two-second debounce, and shares the cross-process sync lock. Polling defaults to five minutes; intervals below 1000 ms fall back to that default.

Include Solidity source files in code search without adding a custom extension.

Since commit 6fc318b
Good to know

Requires the usual embedding and vector database setup; `.sol` is included in the default file extensions.

Before you upgrade Claude Context

Newest first
Show 3 older
v0.1.6

If the Milvus collection has disappeared, `search_code` reports lost index data rather than ordinary empty results and recommends `index_codebase` with `force=true`.

Behavior changedInferredAffects: Users whose local record says indexed but whose database collection is missing.
Read the evidence
Claude Context release history 5
commit 6fc318b

Refresh code search after edits by triggering a sync without waiting for periodic polling.

Details

The MCP server watches `~/.context/.sync-trigger` with a two-second debounce; polling can be disabled or given a custom interval. Branch changes also honor request-specific indexing options, cancel active indexing before clearing its index, and remove overlapping search results.

Also: Control polling, index Solidity, use Gemini Embedding 2, and get corrected file selection, cancellation, and embedding failure reporting.

Build from commit 6fc318b4e3ce58e2898b00a9c3538ead9e24dee5 using pnpm >=10 and Node.js >=20, with Milvus or Zilliz Cloud and a supported embedding provider; these changes are beyond the supplied v0.1.11 tag.

Added: Refresh indexed code on demand from editor hooks an…Added: Include Solidity source files in code search withou…Improved: Generate search embeddings with OpenAI, VoyageAI, G…Now works / inferred: Index a codebase in the background, refresh changed…Now works / inferred: Choose additional file extensions and exclude files…Now works / inferred: Check whether a codebase is indexed and see progres…Behavior changed / inferred: Find relevant code snippets with natural-language q…Action required / inferredBehavior changed / inferred

Seeded by @iluxav with Codex

Read the evidence
v0.1.11
with v0.1.7, v0.1.8, v0.1.9, v0.1.10

Search an indexed project by passing a path inside one of its subdirectories.

Details

search_code and get_indexing_status resolve subdirectory paths to a tracked parent. The grouped tags add OpenRouter and collection-name configuration, skip hidden paths during initial indexing, and fix index recovery and VS Code result navigation.

Also: Add OpenRouter embeddings and Dart indexing; isolate exclusions by project; recover cloud status; run MCP on Node.js 24.

Use @zilliz/claude-context-mcp 0.1.11 or matching core/VS Code packages; Node.js >=20, a supported embedding provider, and Milvus or Zilliz Cloud are required.

Now works / inferred: Find relevant code snippets with natural-language q…Limit lifted / inferred: Find relevant code snippets with natural-language q…Now works / inferred: Check whether a codebase is indexed and see progres…Now works / inferred: Index a codebase in the background, refresh changed…Now works / inferred: Choose additional file extensions and exclude files…Behavior changed / inferred: Choose additional file extensions and exclude files…Added / inferred: Include Dart source files in code search without co…Added / inferred: Generate code-search embeddings using an OpenRouter…Added: Give each codebase's Milvus or Zilliz collection a…Now works / inferred: Generate search embeddings with OpenAI, VoyageAI, G…Now works / inferred: Search and navigate code using the Semantic Code Se…

Seeded by @iluxav with Codex

Read the evidence
v0.1.6
with v0.1.4, v0.1.5

Search an existing vector-database index even when its local indexing record is missing.

Details

search_code checks the vector database when local state is missing and reports a lost collection explicitly. Indexing can recover from stale in-progress state; GEMINI_BASE_URL selects a custom Gemini endpoint.

Also: Recover interrupted indexing and configure a custom Gemini API endpoint.

Use @zilliz/claude-context-mcp 0.1.6 with an existing Milvus/Zilliz setup; custom Gemini endpoints require GEMINI_API_KEY and GEMINI_BASE_URL.

Now works / inferred: Find relevant code snippets with natural-language q…Now works / inferred: Index a codebase in the background, refresh changed…Behavior changed / inferred: Find relevant code snippets with natural-language q…Added: Generate code embeddings through a custom Gemini AP…

Seeded by @iluxav with Codex

Read the evidence
v0.1.3
with v0.1.2

See why background indexing failed and inspect file and chunk counts after it finishes.

Details

get_indexing_status now reports completion counts, failure details, and timestamps. The grouped tags also correct Ollama batch embedding and index clearing, wait for Milvus indexes before loading, and enable syntax-aware Scala and C# chunks.

Also: Ollama batch indexing, clearing index status, Milvus index startup, and Scala/C# syntax-aware chunking are corrected.

Use @zilliz/claude-context-mcp 0.1.3 and matching core/extension packages; Node.js >=20 and <24 and configured embedding/vector database services are required.

Behavior changed / inferred: Check whether a codebase is indexed and see progres…Now works / inferred: Generate search embeddings with OpenAI, VoyageAI, G…Now works / inferred: Index a codebase in the background, refresh changed…Now works / inferred: Find relevant code snippets with natural-language q…

Seeded by @iluxav with Codex

Read the evidence
v0.1.1
Starting point

Find relevant code with natural-language queries from an MCP-compatible coding client.

Details

This baseline summarizes capabilities already present before the ten-tag update window: hybrid code search, background indexing, index management, configurable file selection, and MCP, VS Code, and library access.

Also: Search in VS Code, choose embedding providers, and control which files are indexed.

Use @zilliz/claude-context-mcp 0.1.1 with Node.js >=20 and <24, an embedding provider, and Milvus or Zilliz Cloud; VS Code also has a Semantic Code Search extension.

Added: Find relevant code snippets with natural-language q…Added: Index a codebase in the background, refresh changed…Added: Check whether a codebase is indexed and see progres…Added: Choose additional file extensions and exclude files…Added: Generate search embeddings with OpenAI, VoyageAI, G…Added: Search and navigate code using the Semantic Code Se…Added: Add code indexing and natural-language code search…

Seeded by @iluxav with Codex

Read the evidence