Back to Insights
April 11, 2026AI InfrastructureRob Murtha

What an Agent Can Do with Gerolamo MCP

11 tools, four capability layers, and five ready-made agent configurations. A practical guide to connecting Claude Code, custom agents, or any MCP client to Gerolamo's scored technical intelligence corpus.

Gerolamo is a technical intelligence platform that scores the open-source landscape across GitHub, arXiv, and Hugging Face on defensibility, frontier risk, and novelty. It exposes that intelligence through a REST API and an MCP server — meaning any MCP-compatible agent can query, analyze, and compose from the corpus directly.

This post walks through what an agent can actually do with the Gerolamo MCP tools, how to connect, and five high-value agent configurations you can run today.


1. Discover — "What's out there?"

query_intelligence — Semantic search across all three platforms

A VC analyst's agent asks "what's happening in AI memory systems?" and gets back scored, cross-platform results — repos, papers, and models ranked by relevance.

find_sleepers — Find hidden gems nobody's watching yet

"Find me robotics projects scored 7+ with under 200 stars." Returns repos like a novel SLAM implementation by a solo researcher that's technically excellent but hasn't been discovered.

analyze_competitive_landscape — Map all players in a space

"Show me everything in post-quantum cryptography." Returns 70 entities ranked by velocity — the agent instantly sees who's moving fastest and who's stalling.


2. Analyze — "What does it mean?"

get_intelligence_brief — One-call situation report

An agent runs this every morning for "autonomous agents" and gets: 3 new sleepers, the top trending repo (up 400% this week), and the AI-generated narrative summarizing the state of the space.

get_creator_profile — Research a team or person

"Tell me about deepspeedai." Returns all 5 of their projects, average defensibility of 8.2, total traction of 50k+ stars, authority score indicating they're a proven infrastructure builder.

get_creator_network — Map who works with whom

"Who does huggingface collaborate with?" Returns co-created entities and reach — shows the ecosystem around a creator.

find_defensible_clusters — Find groups of related high-value projects

"Show me clusters of score 7+ projects." Returns groups of entities that share creators or technology — reveals hidden ecosystems.


3. Fuse — "What should I build or invest in?"

compose_molecules — The power tool. Five modes.

Understand mode — "Take these 4 AI memory projects and explain how they relate." Agent gets individual analysis, relationships, and landscape summary. Useful for research prep.

Augment mode — "I have a chatbot platform. How would these 3 MCP server projects enhance it?" Agent gets prioritized integration recommendations, effort sizing, and a numbered task backlog with acceptance criteria. Feed directly into a project management tool.

Build mode — "Fuse this vector database, this embedding model, and this RAG framework into a new product." Agent gets a SPEC.md with architecture, project structure, tech stack, build phases, and copy-pasteable commands. Paste into Claude Code and start building.

Compare mode — "Should I use FalkorDB or Neo4j for my knowledge graph?" Agent gets head-to-head comparison with clear adopt/watch/skip verdicts and risk factors.

Research mode — "Write an investment thesis on the LLM inference serving space using these 5 molecules." Agent gets executive summary, market dynamics, buy/watch/pass on each project, and a strategic thesis. Send to a partner.


4. Monitor — "What changed?"

get_my_latest_intelligence — Check subscription alerts

Agent has an alert for "quantum error correction" — checks it daily and surfaces new papers or repos that crossed the score threshold.

search_intelligence — RAG-synthesized answer

"What is the current state of browser automation AI agents?" Gets a synthesized paragraph drawing from multiple molecules in the corpus — not just a list of results, but an actual answer.


Tool Reference

Tool Returns Use when
query_intelligence Scored molecules with similarity ranking "What exists for X?"
find_sleepers High-score, low-traction molecules "What's under the radar?"
get_intelligence_brief Sleepers + trending + breakouts + narrative "What's happening in X this week?"
get_creator_profile All molecules by a creator + stats "Who is this team?"
compose_molecules (understand) Technology landscape analysis "How do these relate?"
compose_molecules (augment) Integration brief + task backlog "How do I add these to my project?"
compose_molecules (compose) SPEC.md with architecture + commands "Build something new from these"
compose_molecules (compare) Head-to-head with recommendations "Which should I adopt?"
compose_molecules (research) Investment thesis + market analysis "Should we invest in this space?"
analyze_competitive_landscape Velocity-ranked entity list "Who's moving fastest in X?"
get_creator_network Collaboration graph + reach "Who works with whom?"

Connecting to Gerolamo

Option 1: MCP Server (Claude Code / Cursor / Windsurf)

Add to your Claude Code settings (~/.claude/claude_code_config.json):

{
  "mcpServers": {
    "gerolamo": {
      "type": "sse",
      "url": "https://gerolamo.onrender.com/mcp/sse",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Get your API key at gerolamo.org under Settings > API Keys.

Once connected, Claude Code can use all 11 tools directly in conversation:

  • "Search for autonomous drone frameworks scored 7+"
  • "Find sleepers in the robotics space"
  • "Get the intelligence brief for model context protocol"
  • "Compose a Build spec from these entity IDs"

Option 2: REST API (Any Agent)

For custom agents or any HTTP client:

# Semantic search
curl -H "Authorization: Bearer $KEY" \
  "https://gerolamo.onrender.com/api/v1/search?q=autonomous+agents&mode=semantic&limit=10"

# Find sleepers
curl -H "Authorization: Bearer $KEY" \
  "https://gerolamo.onrender.com/api/v1/signals?min_score=7&max_stars=200&limit=10"

# Compose by entity IDs
curl -X POST -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  "https://gerolamo.onrender.com/api/v1/workspace/generate" \
  -d '{
    "mode": "compose",
    "primitives": [...],
    "user_context": "Building a real-time drone coordination platform"
  }'

Compose modes: understand, augment, compose, compare, research

Option 3: CLAUDE.md Integration

For a project that uses Gerolamo as its intelligence source, add to your CLAUDE.md:

## Intelligence Source

This project uses Gerolamo (https://gerolamo.org) for competitive
technical intelligence. The Gerolamo MCP server is connected and
provides these tools:

- `query_intelligence` — search the scored corpus
- `find_sleepers` — discover hidden gems
- `get_intelligence_brief` — daily situation report
- `compose_molecules` — fuse technologies into specs

When researching technologies, libraries, or approaches, query
Gerolamo first to check defensibility scores, threat profiles,
and competitive positioning before making architecture decisions.

Five Agent Configurations

The Morning Scout

Every morning, run get_intelligence_brief for your 5 key topics. If any sleepers or breakouts appear, run compose_molecules in Research mode on them. Email the result to your team.

The Competitive Tracker

For each company in your portfolio, run get_creator_profile weekly. Track their molecule count, average score, and velocity over time. Alert if a competitor's output accelerates.

The Build Agent

Search for a capability you need. Find the top 3-5 molecules. Compose in Build mode. Pass the SPEC.md to Claude Code. Start implementation. Discovery to code in one pipeline.

The Due Diligence Agent

For an acquisition target, run get_creator_profile + compose_molecules in Compare mode against competitors. Get a Research brief on the market. All data-driven, all from the corpus.

The Alpha Hunter

Run find_sleepers daily with min_score=7, max_traction=100. These are the highest-signal, lowest-visibility molecules in the entire corpus. The repos that are about to break out but haven't yet. Track them in bookmarks and check get_intelligence_brief for breakout detection.


Own your intelligence. Build with certainty.