munchbase

name

munchbase

description

General-purpose Crunchbase research agent. Uses the local `munchbase` CLI to run structured discovery, funding, investor, and signal analysis, then synthesizes findings into an evidence-backed briefing. Use when: user asks to research companies, investors, markets, funding trends, founder context, or "what Crunchbase shows" on a topic. Not for: posting data externally, account admin tasks beyond normal auth flow.

Munchbase Research

Research-first workflow over Crunchbase data using munchbase. Focus on:

• clear question decomposition
• high-signal query strategy
• evidence-backed synthesis
• explicit uncertainty and gaps

For endpoint/operator details, see:

• docs/CRUNCHBASE_API_REFERENCE.md
• skills/munchbase/references/munchbase-api.md

Scope

Use this skill when the user needs:

• company profiling
• funding timeline analysis
• investor landscape mapping
• people/founder context
• market/sector discovery
• recent signals and movement

Do not use this skill for:

• social posting or outreach workflows
• broad web-only research where Crunchbase data is not relevant

Research Intake

Before querying, lock these inputs:

• target: company/person/theme/sector
• objective: what decision this research supports
• depth: quick scan vs deep brief
• constraints: geography, stage, time range, categories, output format

If missing, ask concise clarifying questions first.

Research Loop

1. Decompose the question

Split into 3-6 sub-questions, for example:

• who are the relevant entities?
• what happened (funding, signals, key changes)?
• who is involved (investors, founders, operators)?
• how does this compare to peers?

2. Discover entities

Start broad:

• munchbase quicksearch all --query "<topic>" --limit 5

Resolve canonical IDs:

• organization uuid for funding queries
• permalink for profile/entity retrieval

3. Expand into structured queries

Pick only the commands needed for the question:

• profile: org profile, entity get
• funding: org rounds, search funding_rounds
• round participants: round investors
• relationship filtering: subquery search
• semantic discovery: vector search
• activity signals: feed signals
• custom/advanced: search, graph

4. Refine and verify

After each query:

• remove noise with tighter predicates/operators
• add fields required to prove claims
• cross-check important findings with a second query path

5. Synthesize

Group by themes, not command order:

• company state
• funding trajectory
• investor pattern
• notable signals
• risk/uncertainty

Playbooks

Company Brief

1. quicksearch all to identify target entity
2. org profile for baseline profile
3. org rounds for funding history
4. round investors for key rounds if needed
5. summarize trajectory and current status

Investor Landscape

1. identify target company/sector entities
2. pull relevant rounds
3. map recurring investors across rounds
4. cluster by lead vs non-lead participation
5. call out strongest investor patterns

Funding Timeline

1. org rounds with ordering by date
2. capture amount/stage/investor count changes
3. identify acceleration/slowdown periods
4. annotate anomalies and missing data

Competitor/Peer Scan

1. seed with target description via vector search
2. filter peer set with search organizations
3. compare core profile/funding metrics
4. report relative position and caveats

Signal Monitoring

1. run feed signals (news|insight|prediction)
2. filter by categories/locations/lists when provided
3. summarize directional shifts and confidence

Refinement Heuristics

• Too noisy:
  • narrow fields and predicates
  • use exact identifiers where possible
  • reduce scope and increase precision first
• Too sparse:
  • broaden query terms
  • reduce restrictive predicates
  • expand via related entities
• Conflicting indicators:
  • run one independent corroborating query
  • mark confidence as medium/low until resolved

Grounding Rules

Every material claim must map to returned data:

• entity identifier (name/permalink/uuid)
• field(s) used
• observed value(s)

If evidence is partial:

• state what is known
• state what is missing
• propose next query to resolve

Output Style (Required)

• Always end with:
  • Recommended next queries
  • a numbered list of concrete follow-up queries (2-6 items), e.g. tighter filters, excluded round types, investor extraction, or corroboration passes.

Failure Handling

• Auth/session issues:
  • run/redo munchbase auth login
  • confirm with munchbase auth check
• Empty results:
  • broaden query, then re-narrow
  • verify identifiers/permalinks/uuid inputs
• API/shape drift:
  • switch to smaller scoped query
  • capture raw payload with --raw for diagnosis

Limits and Caveats

• Crunchbase web endpoints are undocumented and may change without notice.
• Availability of fields/collections can vary by account/data coverage.
• Treat absence of evidence as "unknown", not "false".