Atlas / MCP servers / lucifer-ux / Contextcore

ContextcoreCAUTION

Name: Contextcore
Author: lucifer-ux

mcp/lucifer-ux/contextcore-2

MCP server by lucifer-ux · ai tools · claude desktop · claude tools · Audited 2026-10-10

ContextCore: An MCP server for Claude (or any AI tool) that enables massive token saving through hybrid search (BM25 + Embeddings)

Verdict

CAUTION

Grade

Trust score

80 /100

Exposed tools

16 15r · 0w · 1d

Transport

—

License

AGPL-3.0

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

One MCP server for all your local files. Search everything, send only what matters to AI.

Stop pasting entire files into Claude. ContextCore indexes your notes, code, documents, images, audio, and video locally — then exposes a single MCP server that any AI tool can query. Instead of bloating your context window, Claude searches first and retrieves only the relevant chunks.

57% fewer tokens. Same answers. No cloud.

Works with: Claude Desktop · Claude Code · Cursor · Cline · OpenCode · any MCP-compatible tool

Why ContextCore?

Most developers working across large codebases or document collections hit the same wall: pasting everything into context is expensive, slow, and hits limits. RAG pipelines require infrastructure. Other memory tools are cloud-only or single-format.

ContextCore is a local-first MCP server that does hybrid search (BM25 + embeddings) across every file type you care about — and registers itself with your AI tools automatically. One install. One server. All your data.

It is not Supermemory. It does not sync to a cloud. Your files stay on your machine. What it gives you is supercharged retrieval across every local file format, surfaced directly inside Claude and other tools via MCP.

This is the one MCP to rule them all design: instead of managing separate MCP servers for different file types, you have one local server with a consistent search API across text, code, images, audi

Read from source at commit 475188e27c9cOBSERVED · 2026-10-10

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add search-embed-sdk --env CONTEXTCORE_GITHUB_TOKEN=${CONTEXTCORE_GITHUB_TOKEN} --env CONTEXTCORE_TEXT_TURSO_AUTH_TOKEN=${CONTEXTCORE_TEXT_TURSO_AUTH_TOKEN} --env CONTEXTCORE_TURSO_AUTH_TOKEN=${CONTEXTCORE_TURSO_AUTH_TOKEN} --env GITHUB_TOKEN=${GITHUB_TOKEN} -- npx -y search-embed-sdk

claude-desktop

{
  "mcpServers": {
    "search-embed-sdk": {
      "command": "npx",
      "args": [
        "-y",
        "search-embed-sdk"
      ],
      "env": {
        "CONTEXTCORE_GITHUB_TOKEN": "${CONTEXTCORE_GITHUB_TOKEN}",
        "CONTEXTCORE_TEXT_TURSO_AUTH_TOKEN": "${CONTEXTCORE_TEXT_TURSO_AUTH_TOKEN}",
        "CONTEXTCORE_TURSO_AUTH_TOKEN": "${CONTEXTCORE_TURSO_AUTH_TOKEN}",
        "GITHUB_TOKEN": "${GITHUB_TOKEN}"
      }
    }
  }
}

Exposed tools (16)

15 read · 0 write · 1 destructive. Blast radius: 1 tool can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`analyze_code_directory`	read
`fetch_content`	read
`filesystem_access_profile`	read
`get_codebase_context`	read
`get_codebase_index`	read
`get_file_content`	read
`get_module_detail`	read
`get_neighbors`	destructive	budget = _consume_budget(session_id=session_id, reset=False)
`index_content`	read
`list_files`	read
`list_sources`	read
`prepare_file_for_tool`	read
`refine_search`	read
`reveal_file`	read
`search`	read
`search_code_chunks`	read

Trust audit

CAUTIONgrade B · trust 80/100 Install with care. The audit found things worth knowing before you trust its output.

Layer	What it checks	Result
L0	Provenance & inventory	PASS
L1	Static analysis of the code	WARN
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (4 observation(s))
Network: declared (5 observation(s))
Shell: declared (1 observation(s))
Dependencies: not all pinned
Secrets in source: none-found