ai-cortex
Local store giving coding agents codebase maps, session history, and typed memories via MCP.
Tested on LinuxKnown vulnerabilities
Use it
Before you install: known vulnerabilities
pickfit found 10 known high or critical vulnerabilities in the packages a fresh install brings (OSV.dev, 3 Nov 2026). Weigh this before you install it.
npm:protobufjs 6.11.6Arbitrary code execution in protobufjs (GHSA-xq3m-2v4x-88gg, critical, fixed in 7.5.5)npm:sharp 0.32.6sharp inherited vulnerabilities in libvips: CVE-2026-33327, CVE-2026-33328, CVE-2026-35590, CVE-2026-35591 (GHSA-f88m-g3jw-g9cj, high, fixed in 0.35.0)
See 8 more
npm:protobufjs 6.11.6— GHSA-66ff-xgx4-vchm (high), fixed in 7.5.6npm:protobufjs 6.11.6— GHSA-685m-2w69-288q (high), fixed in 7.5.6npm:protobufjs 6.11.6— GHSA-75px-5xx7-5xc7 (high), fixed in 7.5.6npm:protobufjs 6.11.6— GHSA-jvwf-75h9-cwgg (high), fixed in 7.5.6npm:protobufjs 6.11.6— GHSA-wcpc-wj8m-hjx6 (high), fixed in 7.6.1npm:sharp 0.32.6— GHSA-rgj7-g3m4-5g8c (high), fixed in 0.35.4npm:sharp 0.32.6— GHSA-wq5f-xc86-pv6w (high), fixed in 0.35.5npm:zod 3.25.76— GHSA-mrhx-c7xc-m3w5 (high), fixed in 4.6.6
Linux x86_64
You need first
- Runtimes:
Node.js 20 or newer
install ai-cortex via npm
npm install -g ai-cortex
Where the README and reality differ
- README: MCP tools listed as recall_memory(worktreePath), record_memory(worktreePath), index_project etc
What worked: actual MCP schemas use `path` for index_project and `task` for suggest_files (probe truncated names); calling with worktreePath/query gives Invalid arguments
Check that it works
ai-cortex --versionThe output should contain “ai-cortex 0.18.1”.
Add it to your agent
claude mcp add ai-cortex -- ai-cortex mcpJSON for Claude Desktop, Cursor and other clients
{
"mcpServers": {
"ai-cortex": {
"command": "ai-cortex",
"args": [
"mcp"
]
}
}
}Add ai-cortex mcp as an MCP server in your coding agent's settings, then use its tools from the agent.
What it does
- Store typed project memoriesTested
How to use it
- Connect your MCP client to
ai-cortex mcp. - Call the
record_memorytool withworktreePathset to <your repo path>,type(for exampledecision), atitleand abody.
- Connect your MCP client to
- Index a git repositoryNot tested
- Rank the files most relevant to a taskNot tested
- Search earlier agent sessionsNot shown working
How far pickfit got
Linux x86_64
- Installs
- Connects
- 1 of 4 features work
- Needs the network
Good for
- index a git repo and rank relevant files for a task
- search prior agent sessions for decisions and corrections
- store and recall typed project memories like decisions and gotchas
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
Linux x86_64
An MCP client called one of its tools
- Replayed from scratch
- Needed the network
- Checked on 31 Oct 2026
- Version 0.18.1
An MCP client handshake over stdio listed 39 tools, and a call to search_history returned a result.
Differences the test machine hit
- README: ai-cortex index /path/to/repo
In the test: CLI index works; MCP index_project also requires `path` not `worktreePath` and is sensitive to cwd