MSDS Chain MCP Server

Chemical safety intelligence for AI-assisted experiment design.

Powered by ChainSDS — a verified, always-current chemical safety database. Verified. Current. Growing.

An MCP server that gives AI agents (Claude Code, Cursor, Copilot, pi, etc.) access to chemical safety reasoning — compatibility checks, hazard analysis, regulatory compliance, PPE recommendations, storage guidance, and more.

Built for researchers who design experiments with AI and need safety verification integrated into their workflow.

Why This Exists

When you use Claude to plan a synthesis route or set up an Opentrons protocol, safety validation shouldn't be a separate step. This MCP server lets your AI assistant automatically:

• Check if chemicals on the same deck are compatible

• Flag dangerous mixing orders

• Recommend PPE for the specific chemicals you're handling

• Verify compliance with EU REACH, US OSHA/TSCA, and 7 other jurisdictions

• Generate signed audit reports for GLP/GMP compliance

Related MCP server: Clarity MCP

Tools (23)

Quick Start

1. Get an API key

Sign up at msdschain.lagentbot.com → API Keys tab → create a key.

2. Install

git clone https://github.com/littleblakew/msds-chain-mcp.git
cd msds-chain-mcp
pip install -r requirements.txt

3. Add to your AI coding agent

Architecture note: the hosted endpoint https://mcp.lagentbot.com sits behind a distribution gateway that terminates OAuth 2.1 (PKCE + DCR) and serves /.well-known/oauth-authorization-server. The OAuth / browser sign-in flow described below applies only to that hosted gateway. This repo's bare core server (python server_remote.py, or the Docker image) does not implement OAuth or serve any /.well-known endpoint — it only accepts Authorization: Bearer sk-msds-... or X-Api-Key passthrough. If you self-host the core directly, use a static API key; there is no browser sign-in step.

Claude Code (Remote — recommended):

claude mcp add msds-chain --transport http https://mcp.lagentbot.com/mcp

Then run /mcp in Claude Code and authenticate — a browser opens to sign in (email code); your account is provisioned automatically and calls are metered to you (free plan includes a monthly free quota). The legacy SSE endpoint (--transport sse https://mcp.lagentbot.com/sse) still works but streamable HTTP is preferred.

Claude Code (Plugin — includes skill + MCP):

/plugin install https://github.com/littleblakew/msds-chain-mcp.git

Claude Code (npm — remote, shim registers the hosted endpoint):

claude mcp add msds-chain -- npx -y msds-chain-mcp@latest

claude.ai (Web): Search "msds-chain" in Settings > Plugins (already published).

Manual config (Claude Code ~/.claude.json):

{
  "mcpServers": {
    "msds-chain": {
      "type": "http",
      "url": "https://mcp.lagentbot.com/mcp"
    }
  }
}

Restart Claude Code, run /mcp, and authenticate (OAuth). You should then see msds-chain in the MCP tools list. (Prefer a static key instead of OAuth? Pass it as a header — --header "Authorization: Bearer sk-msds-your-key" — for headless clients.)

Claude Code Skill

The /msds-safety-check skill provides auto-detection and guided audit workflows.

Plugin install (includes skill + MCP automatically):

/plugin install https://github.com/littleblakew/msds-chain-mcp.git

Manual install (skill only, if MCP already configured):

git clone https://github.com/littleblakew/msds-chain-mcp.git /tmp/msds-chain-mcp
cp -r /tmp/msds-chain-mcp/skills/msds-safety-check .agents/skills/msds-safety-check
ln -s ../../.agents/skills/msds-safety-check .claude/skills/msds-safety-check

What the Skill Does

• Auto-detects chemicals in your conversations and offers safety checks

• /msds-safety-check — guided audit for lab protocols or EHS compliance

• Freemium — basic queries work without API key, audit reports require free registration

Usage Examples

Experiment Protocol Review

User: I'm planning a Grignard reaction with magnesium turnings, diethyl ether,
      and bromobenzene. Check if this setup is safe.

Claude:
  → calls batch_safety_check(["magnesium", "diethyl ether", "bromobenzene"])
  → Returns: compatibility matrix + per-chemical risk warnings (with source SDS)

Opentrons Deck Safety Audit

User: My Opentrons deck has these in different slots:
      Slot 1: Acetone, Slot 3: Concentrated H2SO4, Slot 5: Methanol,
      Slot 7: Sodium borohydride. Any safety issues?

Claude:
  → calls check_chemical_compatibility(["acetone", "sulfuric acid", "methanol", "sodium borohydride"])
  → ⚠️ INCOMPATIBLE: Sodium borohydride + sulfuric acid (violent reaction, H2 gas evolution)
  → ⚠️ CAUTION: Acetone + sulfuric acid (exothermic)
  → Recommendation: Move sodium borohydride to maximum distance from acids

Compliance Check Before Shipping

User: We need to ship toluene and dichloromethane to our Japan lab.
      What transport regulations apply?

Claude:
  → calls get_transport_classification(["toluene", "dichloromethane"])
  → calls check_regulatory_compliance(["toluene", "dichloromethane"], regions=["JP"])
  → Returns: UN numbers, IATA packing instructions, Japan-specific regulations

Generate Signed Audit Report

User: Create a safety audit report for our quarterly review.
      Chemicals: acetone, methanol, ethanol, isopropanol, hexane.

Claude:
  → calls create_audit_session("Q2 2026 Solvent Cabinet Review", ["acetone", "methanol", "ethanol", "isopropanol", "hexane"])
  → calls get_audit_report("SESSION-ID")
  → Returns: Signed PDF URL (Ed25519 signature, suitable for GLP/GMP compliance)

Try These Prompts (one per tool)

Each prompt reliably triggers the named tool. Use them to evaluate the connector end-to-end.

Privacy & Data Handling

• Privacy Policy: msdschain.lagentbot.com/privacy

• Read-only tools send only the chemical names / queries you provide; no personal data is required.

• API-key tools (upload_msds_pdf, create_audit_session, get_audit_report) associate activity with your account for audit traceability.

• Uploaded MSDS PDFs are processed for data extraction and contribute to the verified ChainSDS database per your account's data-sharing settings.

Third-Party AI Platform Integration

Connect to the hosted MSDS Chain MCP server from any AI platform that supports MCP.

Server URL (preferred): https://mcp.lagentbot.com/mcp (streamable HTTP)

The hosted endpoint serves both transports — streamable HTTP (/mcp, preferred) and SSE (/sse, for clients that only speak SSE: 悟空, Dify, Coze, etc.). Calls are metered per authenticated user.

Authentication: either OAuth (the client opens a browser sign-in on first connect) or a static header — Authorization: Bearer sk-msds-your-key — for headless clients.

悟空 (Wukong)

1. 设置 → MCP 服务 → + 添加

2. 类型：SSE

3. 名称：msds-chain

4. URL：https://mcp.lagentbot.com/sse

5. HTTP Headers：Authorization : Bearer sk-msds-your-key

6. 点击「添加」

Gemini CLI

Listed in the Gemini CLI extensions gallery. Install it with:

gemini extensions install https://github.com/littleblakew/msds-chain-mcp

Gemini asks you to confirm twice: once to trust the workspace, once to review the extension and the agent skill it carries. Installing it gives you two things:

• the hosted MCP server at https://mcp.lagentbot.com/mcp (remote, nothing runs locally)

• the msds-safety-check skill, which prompts Gemini to run a safety check when you are working through an experimental procedure

Signing in. Your first chemical question starts the sign-in automatically. Gemini opens a browser, you enter your email, and you paste back the code we send you. There is no API key to create beforehand and no credentials to fill in by hand: the server supports Dynamic Client Registration, so Gemini registers itself and completes the OAuth exchange on its own. A free account includes a monthly allowance.

Check that it connected:

gemini mcp list

Then just ask, in chat:

Can acetone and sodium hypochlorite be stored in the same cabinet?

Gemini picks the right tool for the question. Every answer names the supplier SDS it came from, with a link to the original PDF.

Dify / Coze / other platforms

General steps:

1. Find MCP server / tool integration settings

2. Select SSE transport type

3. Set URL to https://mcp.lagentbot.com/sse

4. Add authentication header: Authorization: Bearer sk-msds-your-key

Get an API key: Sign up at msdschain.lagentbot.com → API Keys tab → Create Key.

Remote Mode (HTTP)

For cloud deployment or shared team access, run as an HTTP server:

MSDS_API_KEY=sk-msds-xxx python server_remote.py

A single server_remote.py process always serves both transports concurrently — Streamable HTTP at /mcp (recommended for Claude Code 2026+) and SSE at /sse (compatibility). The MSDS_MCP_TRANSPORT env var is kept for backward compatibility only; it is ignored at runtime.

Connect from Claude Code:

claude mcp add msds-chain --transport http https://your-server.example.com/mcp

Docker

docker build -t msds-chain-mcp .
docker run -p 8080:8080 -e MSDS_API_KEY=sk-msds-xxx msds-chain-mcp

Configuration

Use Cases

Lab Automation (Opentrons / Hamilton / Tecan)

• Pre-run protocol safety audit

• Deck layout compatibility verification

• Automated run compliance reports

Electronic Lab Notebooks (Benchling / LabArchives)

• Safety annotations on experiment entries

• Chemical registration with auto-flagging

Pharmaceutical R&D

• Synthesis route safety screening

• Multi-region regulatory pre-checks for new compounds

• GMP-ready audit trail with signed receipts

Data Coverage — ChainSDS

Industry-sourced, AI-verified, and cryptographically signed.

• 4,000,000+ chemical records with multi-language aliases (EN/ZH/JA)

• NFPA/GHS classification for compatibility rules

• 23 regulatory watch lists across 8 jurisdictions + 3 international conventions (EU, US, CN, JP, KR, CA, AU, SG; no Taiwan and no IARC coverage)

• Occupational exposure limits from 5 standards (OSHA PEL, ACGIH TLV, EU IOELV, JP OEL, CN GBZ)

• UN transport data for 16+ common lab chemicals

• Version tracking — detects a newer SDS and diffs its hazard classification against the one you relied on

Architecture

Your AI Agent                  MSDS Chain MCP Server
┌──────────────────┐           ┌─────────────────────────┐
│ Claude Code      │           │ 23 Safety Tools         │
│ Cursor / pi      │──MCP────▶│   ↓                     │
│ Any MCP client   │           │ ChainSDS Platform       │
└──────────────────┘           └─────────────────────────┘

The MCP server connects to the ChainSDS platform for verified safety reasoning.

Development

Test locally with the MCP inspector:

export MSDS_API_KEY=sk-msds-your-key
npx @modelcontextprotocol/inspector python server.py

Troubleshooting

Still stuck? Email contact@lagentbot.com or open an issue at github.com/littleblakew/msds-chain-mcp/issues.

Roadmap

• get_waste_disposal — waste classification and disposal guidance

• check_mixing_order — safe addition sequence for reagent pairs

• get_chemical_alternatives — safer substitutes for restricted chemicals

• Remote MCP (HTTP SSE / Streamable HTTP) for cloud-hosted access

• OAuth 2.1 for Claude Marketplace integration (served by the distribution gateway in front of mcp.lagentbot.com, not by this core server)

License

MIT

About

Built by LAgentBot — AI-powered chemical safety infrastructure.

Part of the MSDS Chain platform — the world's first AI Agent-driven chemical safety data trust network, powered by ChainSDS: verified, current, and growing.