chunk-invariance MCP server
io.github.ninadphalak/chunk-invariance·v0.2.0·13 starsChecks whether your streaming redactor leaks a secret split across chunks; writes a pytest for it.
13 stars54 downloads/wk
Reviews
Write oneNobody has reviewed chunk-invariance yet.
If you have run it, two minutes of your experience saves the next person an afternoon.
chunk-invariance tools (3, 1 write)
write = sends, deletes, buys or postsRead from the package source without running it. The installed server may list more.
check_stream_filterexplainExplain in one paragraph why filtering a stream chunk by chunk leaks values cut across chunks, with an example and the fix. No inputs.
write_regression_testwrite action
Public scan report
scanner v0.1.10 · 2026-10-27 · same rubric, same numbers if you re-run it
- Code scan9 source files scanned25/25
- –Live reliabilityno gateway calls yet and no remote to proben/a
- –Tool poisoningtools not inspected (local package is not executed); not countedn/a
- Auth qualitylocal package, no credentials required12/15
- Maintenancelast push 2 days ago15/15
- Maintainer identityregistry namespace matches repository owner; GitHub account older than a year8/10
How chunk-invariance compares
Against every graded Security MCP server, from our own scans and uptime checks. Updated after each scan run.
- Safety score
- 92/100
- higher than 95% of 918 Security servers
- Tools
- 3
- the typical Security server has 6
- Tools that change things
- 33%
- typical: 0%. Fewer write tools means less an agent can break
- GitHub stars
- 13
- more than 91% of 585 Security servers
- Downloads a week
- 54
- more than 47% of 240 Security servers with a package
What its tools let an agent do: 2 read · 1 create
Install chunk-invariance in Claude Code, Cursor or VS Code
claude mcp add chunk-invariance -- uvx chunk-invariance
What the publisher says
From the chunk-invariance repository's README, as published. We do not edit it. Read it on GitHub
LLM-Shield-Proxy
LLM-Shield-Proxy is a self-hosted gateway for OpenAI-compatible LLM APIs. It replaces the personal data and secrets it detects (emails, card numbers, SSNs, API keys and more) before a request goes to the model provider, and puts the original values back into the streamed response before your application sees it. Your application changes only its base_url and the key it sends.
Already running LiteLLM? From LiteLLM 1.106.0 (released 2026-10-18) the Shield is a built-in guardrail, guardrail: llmshieldproxy: LiteLLM keeps its own routing and calls the Shield to redact each request and restore each reply. See LiteLLM's setup page. An older LiteLLM does not stop on that config: it logs one error line, Skipping guardrail 'llm-shield': invalid configuration, proxy is starting WITHOUT this guardrail: Unsupported guardrail: llmshieldproxy, starts, and every request reaches the provider unredacted. Check that line is absent from the startup log, or on an older LiteLLM use one of the wirings in Running behind LiteLLM.
Testing a gateway you already run, rather than adding this one? The benchmark checks whether it sends personal data to its model provider, with no model account and no install of this proxy.
Try it in a minute, with no API key
pip install llm-shield-proxy
UPSTREAM_BASE_URL=http://127.0.0.1:8765 UPSTREAM_API_KEY=unused VALID_VIRTUAL_KEYS=sk-demo llm-shield-proxy --port 4000 &
pii-leak-benchmark selfcheck --target-base-url http://127.0.0.1:4000/v1 --target-api-key sk-demoThe second command starts the proxy. The third sends prompts full of synthetic emails, SSNs, card numbers and API keys through it, and plays the model provider at 127.0.0.1:8765, so it can see exactly what the proxy forwarded. Nothing calls a real model. You should see:
CLEAN
No fixture value reached the upstream; required restore checks passed.
Data types tested
TYPE RESULT WHAT IT MEANS
AWS_ACCESS_KEY_ID contained never reached the upstream in this run
CREDIT_CARD contained never reached the upstream in this run
EMAIL contained never reached the upstream in this run
GITHUB_TOKEN contained never reached the upstream in this run
SLACK_TOKEN contained never reached the upstream in this run
SSN contained never reached the upstream in this runRun the same check with nothing in the middle (--target-base-url capture://self) and every row reads LEAK. On Windows PowerShell, use the PowerShell version of these commands.
Use it with your application
Shortened. The full README is on GitHub.
Nothing above is checked by us. What we check is on the safety report.
chunk-invariance: common questions
- Is chunk-invariance MCP server safe?
- Yes, by our scan: it is graded A (92/100). Read the chunk-invariance safety report
- How do I install chunk-invariance?
- It runs on your machine. Copy the Claude Code, Cursor, VS Code or Claude Desktop config from the install section.
- Does chunk-invariance need an API key?
- Not as far as the registry entry and our scan can tell: no credentials are declared or required.
- Is chunk-invariance maintained?
- The last commit was 7 days ago (2026-10-26). The latest release is v0.2.0.
- What can I use instead of chunk-invariance?
- Servers from other publishers that do the same job: Zn Gate MCP server, Jikida MCP server and Kastell MCP server. Compare all chunk-invariance alternatives.
- How does chunk-invariance compare with other Security MCP servers?
- safety score 92/100 (higher than 95% of 918 Security servers); tools 3 (the typical Security server has 6); tools that change things 33% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.
Alternatives to chunk-invariance
Same job from other publishers: the closest match first, then the best rated.
- Zn GateThe local security layer for coding agents: secret redaction, injection checks and an MCP proxy.not reviewedGrowingB
- JikidaSecurity tools for your AI: scan, pentest, check headers, guard code and scan repos for secrets.not reviewedGrowingA
- KastellServer security audit (413 checks), hardening, and fleet management across 4 cloud providers.not reviewedGrowingB
- SkylosDead code, security, secrets detection and code quality for Python, TypeScript, Go.not reviewedEstablishedC
- Prodcheck4,685 pre-production checks: security, performance, scale, integrations, post-launch, people.not reviewedGrowingA