Sieve Mcp

io.github.lmwhartonPublished by GitHub account lmwhartonv0.1.4Updated Oct 28, 2026

AI startup due diligence — screen companies across 7 dimensions, get a Sieve Score.

VerifiedSTDIODesktop onlyFinanceData & AnalyticsBusiness & Commerce

Overview

AI-generated overview

Sieve MCP lets an assistant screen startups for venture capital due diligence, scoring them 0-140 across seven dimensions and returning an investment memo.

What you can do
  • Start a screening run for a company by name, website, pitch deck text or description
  • Poll analysis progress and see which dimensions are complete with current scores
  • Retrieve the full investment memo with Sieve Score, recommendation, strengths, concerns and evidence
  • Check how many screens have been used in the current billing period
When to use it
Useful for investors, angels, fund managers and accelerators who want a fast first-pass screen of many deals, a structured memo for an investment committee, or a consistent framework to compare startups across a portfolio.
Requirements
Runs locally as a Python package (pip install sieve-mcp or uvx sieve-mcp) over stdio. Requires a Sieve API key supplied in the SIEVE_API_KEY environment variable; an optional SIEVE_API_URL can override the API base URL. Network access to the Sieve service is needed.
Before you install
The API key is a secret passed via SIEVE_API_KEY and should be kept out of shared configs. Starting a screen consumes a paid quota (free tier limited to 2 screens per month, Pro unlimited), and company details, website URLs or pasted pitch deck text are sent to a third-party service for analysis. Screening writes new analysis records on the remote service.

Installation

In SourceWeft

  1. Open Sieve Mcp in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

[PyPI version] [Python 3.10+] [License: MIT] [MCP Compatible]

Sieve MCP Server — AI-Powered Venture Capital Due Diligence

The first MCP server purpose-built for venture capital. Drop a company name into Claude, Cursor, or Windsurf and get a quantified investment memo in 5 minutes — not a ChatGPT summary, a real analyst-grade IMPACT-X assessment with every claim verified against evidence.

"Screen a startup called Acme Corp" → Sieve researches the company, scores it across 7 dimensions, verifies every claim, and tells you: Take the meeting or Pass.

Why Sieve?

Most VCs screen 50+ deals a month. Most of those are obvious passes — but you still spend 2-3 hours per deal on basic diligence before you know that. Sieve does that work in 5 minutes.

What You Get

CapabilityDetails
Investment MemoFull analyst-grade memo with executive summary, key strengths, critical concerns, and actionable recommendation — not a ChatGPT summary
Deal Scoring (0-140)Quantified Sieve Score across 7 IMPACT-X dimensions — structured, repeatable, comparable across your portfolio
Red Flag DetectionAutomatically surfaces deal-breakers: founder risk, market timing issues, unsustainable unit economics, missing traction, competitive threats
Evidence VerificationEvery finding tagged as Documented, Discovered, Inferred, or Missing — cross-references pitch deck claims against real-world evidence
Competitive LandscapeMaps competitors, identifies moat strength, evaluates defensibility and switching costs
Financial Model AssessmentEvaluates unit economics, pricing power, revenue model viability, and path to profitability
Market SizingTAM/SAM/SOM analysis with timing assessment — is the market ready?
Team EvaluationFounder-market fit, team completeness, domain expertise, execution track record
Traction AnalysisGrowth metrics, customer validation, revenue signals, engagement patterns
Real-time Deal ChatAsk follow-up questions, challenge findings, explore what-if scenarios, dig into any dimension
Sector-Aware AnalysisAdapts benchmarks for fintech, healthtech, deeptech, climate, SaaS, consumer, and more
Stage-CalibratedDifferent expectations for pre-seed vs seed vs Series A — doesn't penalize early-stage for missing late-stage metrics

What Makes Sieve Different

  • Not a wrapper around ChatGPT. Sieve runs a structured multi-agent analysis pipeline with real-time web research, evidence verification, and domain-specific benchmarks.
  • We tell you what we don't know. Every finding is evidence-typed (Documented / Discovered / Inferred / Missing) so you see exactly what's verified and where the gaps are.
  • Repeatable framework. The same IMPACT-X methodology every time, so you can compare deals apples-to-apples across your portfolio.
  • Built by VCs. Not a generic AI tool adapted for investing — purpose-built for the way investors actually evaluate deals.

IMPACT-X Framework

DimensionWhat Sieve evaluates
I — InnovatorsFounding team quality, experience, domain expertise
M — MarketOpportunity size, timing, tailwinds
P — ProductSolution strength, differentiation, technical depth
A — AdvantageCompetitive moat, defensibility, switching costs
C — CommerceBusiness model, unit economics, pricing power
T — TractionGrowth metrics, validation signals, customer evidence
X — X-FactorUnique qualities, timing advantages, intangibles

Each dimension scores 0-20. Total Sieve Score ranges from 0-140.

Quick Start

Install

bash
pip install sieve-mcp

Or run directly without installing:

bash
uvx sieve-mcp

Get Your API Key

  1. Sign up free at app.sieve.arceusxventures.com
  2. Go to Settings → copy your API key
  3. Free tier: 2 screens/month. Pro: unlimited.

Claude Desktop

Add to your claude_desktop_config.json:

json
{  "mcpServers": {    "sieve": {      "command": "uvx",      "args": ["sieve-mcp"],      "env": {        "SIEVE_API_KEY": "your-api-key"      }    }  }}

Claude Code

bash
claude mcp add sieve -- uvx sieve-mcpexport SIEVE_API_KEY="your-api-key"

Cursor / Windsurf

Add to your MCP settings:

json
{  "mcpServers": {    "sieve": {      "command": "uvx",      "args": ["sieve-mcp"],      "env": {        "SIEVE_API_KEY": "your-api-key"      }    }  }}

Available Tools

ToolWhat it doesRead-only
sieve_screenStart a Quick Screen — pass a company name, optional website URL, pitch deck text, or descriptionNo
sieve_statusPoll analysis progress — see which dimensions are complete and current scoresYes
sieve_summaryGet the full investment memo — Sieve Score, recommendation, strengths, concerns, evidenceYes
sieve_usageCheck how many screens you've used this billing periodYes

Example Workflow

Just talk to your AI assistant naturally:

1. Screen a startup

"Run a Sieve screen on Acme Corp at acme.com"

2. Check progress (analysis takes 2-5 minutes)

"What's the status of that Sieve analysis?"

3. Get the full investment memo

"Show me the Sieve results for Acme Corp"

4. Explore the deal

"What are the red flags? Anything that should kill this deal?"

"How strong is their competitive moat? Who are the main competitors?"

"Walk me through their unit economics — is this business model viable?"

"How does the founding team stack up? Any gaps?"

"What's the market timing like? Is this too early or too late?"

"Compare this deal to the last three I screened"

5. Check your usage

"How many Sieve screens do I have left this month?"

Pro tip: Upload a pitch deck for deeper analysis

Paste pitch deck text or founder meeting notes for maximum accuracy:

"Screen Acme Corp — here's their pitch deck text: [paste]. Also check acme.com"

Sieve cross-references every pitch deck claim against real-world evidence. When a founder says "fastest-growing in the category," Sieve checks if that's actually true.

More things you can ask

Use caseExample prompt
Quick pass/take meeting"Screen this company and give me the bottom line"
Investment memo for IC"Generate a full investment memo I can present to my partners"
Red flag check"What are the deal-breakers for this startup?"
Competitive analysis"Who competes with this company and how defensible are they?"
Financial viability"Does their business model make sense? What are the unit economics?"
Team assessment"Evaluate the founding team — any gaps or concerns?"
Market validation"Is the market real? What's the TAM and timing?"
Batch screening"Screen these 5 startups and rank them by Sieve Score"
Follow-up prep"Based on this analysis, what questions should I ask the founders?"
Portfolio comparison"How does this deal compare to others I've screened?"

Who Uses Sieve

  • Solo GPs screening 50+ deals/month — stop spending hours on obvious passes
  • Angel investors evaluating startups outside their domain — Sieve brings sector expertise
  • Emerging fund managers building a repeatable diligence process — consistent framework, every time
  • Accelerators standardizing evaluation — compare apples to apples across your cohort

Environment Variables

VariableRequiredDefaultDescription
SIEVE_API_KEYYes—Your Sieve API key (get one free)
SIEVE_API_URLNohttps://api.sieve.arceusxventures.comAPI base URL

Running with Docker

bash
docker build -t sieve-mcp .docker run -p 8080:8080 -e SIEVE_API_KEY=your-key sieve-mcp

Development

bash
# Install locallypip install -e .
# Run in stdio mode (for MCP clients)sieve-mcp
# Run in HTTP mode (for remote/container deployment)sieve-mcp http

Also Available As

Sieve isn't just an MCP server. Use it however fits your workflow:

Links

License

MIT

Built by ArceusX Ventures — from VCs, for VCs.

Source: README.md at commit 635b3cb

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v0.1.4LatestSep 16, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts