VCAgents MCP Server: Octagon-Powered VC Diligence

Overview

This MCP (Model Context Protocol) server runs programmable, agentic VC simulations that combine LLM-driven reasoning with Octagon Private Markets’ live deal data, valuations, and research. Developers can host the server locally or inside a private environment and expose it to tools that support MCP (for example Claude Desktop, OpenAI Agents SDK, or custom agent runners). The server exposes a set of investor personas (simulated VC brains) for tasks such as pitch feedback, diligence simulations, term-sheet negotiation, and portfolio mapping.

Purpose: give teams a lightweight, open-source way to spin up “VC agent” endpoints that are continuously enriched by Octagon data and driven by your OpenAI (or compatible) API key. Use it for rehearsals, founder prep, automated deal triage, or internal tooling that needs structured, investor-style analysis.

Features

• Pre-built VC persona agents for different investing philosophies (e.g., product-first, network-effect, marketplace analytics)
• Live enrichment from Octagon Private Markets (requires an Octagon API key)
• LLM-powered analysis using your OpenAI API key (or other supported LLM backends)
• Exposes an MCP-compatible server so other tools can invoke agents programmatically
• Configurable prompts, tools, and response formats for embedding into workflows
• Local-first: run in an isolated environment (pipx/virtualenv) or containerize for deployment
• Example prompt templates for deal critique, diligence checklists, and outreach drafts

Installation / Configuration

Prerequisites:

• Octagon API key (Octagon Private Markets)
• OpenAI API key (LLM access)
• Python 3.10+ (recommended) and pipx or a virtualenv

1. Clone the repo

git clone https://github.com/OctagonAI/octagon-vc-agents.git
cd octagon-vc-agents

2. Install and run with pipx (isolated, no global install)

# install via pipx from git
pipx install --spec git+https://github.com/OctagonAI/octagon-vc-agents.git octagon-vc-agents

# run the MCP server (example)
OCTAGON_API_KEY="sk_octagon_..." OPENAI_API_KEY="sk_openai_..." \
  pipx run octagon-vc-agents serve --port 3333

3. Or run from a virtualenv

python -m venv .venv
source .venv/bin/activate
pip install -e .
export OCTAGON_API_KEY="sk_octagon_..."
export OPENAI_API_KEY="sk_openai_..."
# example entrypoint
python -m octagon_vc_agents.server --port 3333

4. Quick health check (once server is running)

curl http://localhost:3333/health
# expected: {"status":"ok","service":"octagon-vc-agents"}

Configuration notes:

• Set OCTAGON_API_KEY for Octagon data enrichment
• Set OPENAI_API_KEY (or equivalent) for LLM calls
• Port and logging level can be adjusted via CLI flags or environment variables supported by the package

Available Resources

Agents included (samples — these are simulations, not endorsements):

API surface (typical endpoints):

• /health — readiness probe
• /agents — list available agent personas and metadata
• /agents/{agent_id}/prompt — submit a prompt or dossier to an agent
• /agents/{agent_id}/tools — available tools/enrichment for the agent

(Use the /agents endpoint to inspect agent capabilities and expected request shape.)

Use Cases & Examples

1. Deal critique (copyable prompt)

• Ask two agents for comparative feedback on a company.
• Example: send your 10-slide deck URL and metrics; request a side-by-side table with risks, defensibility, and top 3 questions each investor would raise.

2. Pre-call qualification

• Run an agent to produce a “meet/pass” decision with three metrics to improve before a partner vote.
• Input: one-sentence pitch + link to deck + current traction numbers.

3. Diligence simulation (14-day lens)

• Ask an agent to list missing evidence and tests to run in the next two weeks that would materially change a term-sheet decision.

4. Portfolio intro mapping

• Provide portfolio list + target theme and ask an agent to identify warm intros and draft first-contact emails for each CEO.

Example HTTP call to invoke an agent (JSON body shape is illustrative):

POST /agents/octagon-fred-wilson-agent/prompt
Content-Type: application/json

{
  "prompt": "Evaluate this pre-seed startup: