How it works
Question → agent → DataGardener MCP → evidence → answer.
How an AI agent uses DataGardener MCP: it understands the question, selects intelligence tools, queries entitled modules, connects evidence on company registration number and returns an answer with source and date.
- 1
The host sends the question and the tool list to the model
When a person asks a question in an MCP host such as Claude, the host includes the DataGardener MCP tool definitions in the model's context. The model now knows which tools exist, what arguments they take and what they return.
- 2
The model plans
It decomposes the question into conditions (geography, sector, financial signal, contract history) and maps each to the lightest tool that can evidence it. Describe, count and knowledge-base tools are used first so scope is known before records are pulled.
- 3
The host calls DataGardener MCP
Each tool call travels to the DataGardener MCP server over the authorised OAuth connection. Identity is bound to the connection, so the server applies that customer's entitlement and default filters, for example excluding dissolved companies.
- 4
DataGardener returns structured results
Records come back with their fields, the company registration number, and where applicable the source, source date and usage consumed. Errors come back as readable messages the model can act on.
- 5
The model connects the evidence
Results from different modules are joined on registration number. A company that fails a condition drops out with the reason recorded. The model may call more tools if a gap appears.
- 6
The answer, with its working
The person receives a direct answer, the supporting records or aggregates, the provenance chips and suggested next questions. Nothing is asserted that the evidence does not support.
What the person sees at the end
A short answer, the records, and this:
Watch the stages
The same five stages, animated.
Choose a question
Illustrative output. Real tool names.
Question
Which manufacturing businesses around Birmingham may be approaching a refinancing opportunity?
Stage 1 of 5
Understanding the request
The agent breaks the question into the conditions it must satisfy.
- Geography: Birmingham and surrounds
- Sector: manufacturing
- Charge age
- Lender relationship
- Financial trend
Each condition maps to one or more DataGardener intelligence layers. Nothing is guessed: if a condition cannot be evidenced, the agent says so.
- Company Intelligence
- Lending
- Financial Intelligence
- Risk
- Ownership
Read this investigation as text
Question: Which manufacturing businesses around Birmingham may be approaching a refinancing opportunity?
Intent: Geography: Birmingham and surrounds; Sector: manufacturing; Charge age; Lender relationship; Financial trend.
Tools selected: execute_query, search_ukgaap, search_psc.
Modules queried: Company index (Manufacturing SIC codes within Birmingham postcode districts, live: 3,912 candidates); Company index (charges) (Charge created 36 to 60 months ago, not satisfied, no newer charge: 318 match); UK GAAP financials (Turnover stable or growing, fixed assets rising: 142 match); PSC index (No change of control in last 24 months: 131 shortlisted).
Evidence connected: Charge date and charge holder from the charges register; Fixed-asset growth from the balance sheet; Ownership stability from the PSC register.
Answer: 131 Birmingham-area manufacturers hold a secured facility between three and five years old with no newer charge, stable ownership and rising fixed assets. These are the conditions under which a refinancing conversation is most often welcome.
Provenance: Source Companies House charges and accounts; PSC register; date Charges to Oct 2026; accounts to 2025; module Lending, Financial, Ownership; usage 131 company records.
Suggested next questions: Rank by total secured debt / Which have asset-finance charges specifically? / Build a borrower brief for the top five
Your agents already know how to think.Give them something worth knowing.
Built on DataGardener intelligence already used by leading UK organisations.