Electronics Rfq Agent

Electronics RFQ Agent

Your sales engineers are spending 2-4 hours turning RFQ documents into quotes. This does it in seconds.

Electronics RFQ Agent is a Python library and CLI that reads RFQ documents (PDF, Excel, Word), looks up every line item against your ERP catalog, and outputs a draft quote. It connects to SAP, Epicor, Oracle, and Microsoft Dynamics through MCP servers, so it works with Claude, GPT-4, or any agent framework that speaks MCP.

---

Table of contents

• Install
• The problem this solves
• Quickstart
• Commands
• API reference
• How it differs from the alternatives
• ERP support
• Benchmarks
• Integration matrix
• Try it in Docker
• Security
• FAQ
• Contributing
• License

Install

pip install electronics-rfq-agent-cli

or

uv add electronics-rfq-agent-cli

To install from source instead:

git clone https://github.com/RudrenduPaul/electronics-rfq-agent

cd electronics-rfq-agent pip install -e .

or, with uv:

uv sync

Parsing RFQ documents (PDF, Excel, Word) calls the Anthropic API, so set ANTHROPIC_API_KEY before running anything that touches a real document:

export ANTHROPIC_API_KEY="sk-ant-..."

You don't need this key to run erfa audit against an existing quote file, or to explore the CLI with --help. Only document parsing calls out to Claude.

The problem this solves

We were working with electronics distributors who had 3-5 sales engineers spending most of their day on quote entry. Every tool we found was either tied to one specific ERP or required a 6-month integration project. We wanted something that worked with what distributors already had, could be self-hosted (quote data is sensitive), and was actually extensible.

The MCP architecture means adding a new ERP is writing one file. The parser handles the document formats distributors actually send: hand-filled PDFs, multi-sheet Excel files, and the occasional scanned table.

Quickstart

No ERP system required to try it out; the mock backend ships with 200 realistic electronics parts. You do need ANTHROPIC_API_KEY set, since parsing the RFQ document is still a real Claude call:

from electronics_rfq_agent import QuoteAgent

from electronics_rfq_agent.mcp.mock import MockERP

agent = QuoteAgent(erp=MockERP()) quote = agent.run_sync("path/to/rfq.xlsx")

for line in quote.lines: print(f"{line.rfq_line.part_number}: {line.status} @ {line.unit_price}")

print(quote.summary())

MockERP applies quantity-based pricing tiers automatically: qty >= 1000 gets 20% off, qty >= 100 gets 10% off, qty >= 10 gets 5% off. List price applies below qty 10. This mirrors real-world volume pricing so benchmarks and integration tests reflect realistic cost curves.

Connect to a real ERP:

from electronics_rfq_agent import QuoteAgent

from electronics_rfq_agent.mcp import EpicorMCP

agent = QuoteAgent( erp=EpicorMCP(base_url="https://your-epicor.company.com", api_key="..."), max_concurrent=10, # parallel ERP lookups (default: 10) ) quote = agent.run_sync("rfq_2026_0619.xlsx") print(quote.summary())

Commands

erfa ships two subcommands. Every flag below is pulled straight from erfa --help.

# Generate a quote from an RFQ file against the mock ERP

export ANTHROPIC_API_KEY="sk-ant-..." erfa quote rfq.xlsx --mock

Save the quote as JSON for later inspection

erfa quote rfq.xlsx --mock --output quote.json

Audit what happened: what was found, substituted, or missing and why

erfa audit quote.json

Audit output example (real output from erfa audit docs/example-quote.json, generated by running the mock ERP's pricing logic against the sample RFQ in tests/fixtures/sample_rfq.txt):

Audit Report - Quote df9fd083

RFQ Source : tests/fixtures/sample_rfq.txt Lines : 5 Total : USD 64.04

FOUND (4)

---

L 1 RES-0402-10K-1PCT qty=1000 unit=0.0064 ext=6.40 L 2 CAP-100NF-50V-X7R-0402 qty=500 unit=0.0104 ext=5.20 L 3 IC-LM358-SOIC8 qty=50 unit=0.7101 ext=35.50 L 4 XTAL-16MHZ-SMD qty=25 unit=0.6774 ext=16.94

NOT FOUND (1)

---

L 5 MOSFET-NMOS-20V-3A-SOT23 Part 'MOSFET-NMOS-20V-3A-SOT23' not found in ERP catalog

Fill rate: 80% (4 found / 0 substituted / 1 not found)

Zero-price parts: If a part exists in the ERP catalog but has a unit price of $0.00, the agent quotes $0 rather than skipping the line, and sets line.notes to a message flagging the zero price so you catch it before quoting the customer. Check line.notes for any found or substituted line before sending a quote out.

API reference

The full reference lives in docs/api.md: every QuoteAgent parameter, the shared ERP connector interface, Quote/QuoteLineItem field-by-field, and the exception hierarchy. The exports below are what from electronics_rfq_agent import ... actually gives you, grepped from src/electronics_rfq_agent/__init__.py, not guessed:

How it differs from the alternatives

ERP support

SAP note: pyrfc requires the SAP NetWeaver RFC Library, which is not on PyPI and must be downloaded manually from SAP's support portal (S-user required). See docs/erp-setup/sap.md for step-by-step instructions.

Benchmarks

Measured using the in-memory mock backend (200 realistic parts, no ERP system required). Run it yourself:

git clone https://github.com/RudrenduPaul/electronics-rfq-agent

cd electronics-rfq-agent uv run python benchmarks/run.py

ERP lookup latency (100 individual lookups, mock backend):

Quote assembly time by RFQ size (parser mocked, no Anthropic API call):

*These numbers cover ERP lookup and quote assembly only. The RFQ parser is mocked and no Anthropic API call is made. In production, AI document parsing adds 5-15s per document and real ERP lookups add 100-500ms per line (parallelised at up to max_concurrent=10), giving a realistic total of ~15s for a 50-line RFQ. Manual baseline: 2-4 hours.*

Integration matrix

Electronics RFQ Agent works with any agent framework that supports MCP:

Try it in Docker

git clone https://github.com/RudrenduPaul/electronics-rfq-agent

cd electronics-rfq-agent cp .env.example .env # add your ERP credentials and Anthropic API key docker compose up -d

Your quote data never leaves your environment.

Security

• Vulnerability scanning: Trivy scans every CI run (HIGH/CRITICAL severity, build fails on unfixed findings), results uploaded to the GitHub Security tab.
• Release signing: the release workflow signs built wheels and sdists with Sigstore and attaches a CycloneDX SBOM to each GitHub Release once a version is tagged.
• Dependency pinning: Dependabot keeps GitHub Actions and Python dependencies current (weekly patch updates, monthly Actions updates).
• Disclosure: SECURITY.md. Report vulnerabilities privately via email or GitHub Security Advisories; 48-hour acknowledgement, 90-day resp