RAN Intelligence MCP · Paid agent access
Give your AI the commercial roofing record.
Connect an approved AI workspace to RAN’s source-backed Roof IDs, Bid Truth pursuits, changes, project evidence, operator decisions and customer-approved private context. Every result stays read only, tenant bound and traceable to the evidence behind it.
What it does
Answers operators can act on.
This is not a public data dump and it is not another chatbot. RAN normalizes official sources, preserves revision history, applies the operator’s approved priorities and returns the evidence an agent needs to explain its answer.
What changed?
Compare canonical source snapshots, detect revisions and material movement, and distinguish corroboration from contradiction.
Why does it matter to us?
Rank evidence against the organization’s territory, watchlist and approved learning profile—without silently changing policy.
What did we decide?
Retrieve permanent Decision Receipts with owner, deadline, rationale, source lineage, outcome and verification fingerprint.
Access flow
Five steps. Human control stays intact.
The agent can investigate and retrieve. It cannot approve a brief, change a project, rewrite a decision or expand its own access.
Activate
RAN approves the organization and enables the paid MCP entitlement.
Issue
An authorized operator creates a dedicated token in Agent Access.
Connect
Add the RAN endpoint and bearer token to a compatible MCP client.
Investigate
The agent retrieves scoped evidence, sources, changes and decision history.
Approve
People review actions and publisher outputs inside the Operator Desk.
Reference account
Proven with Division 7.
Division 7 is the founding operator reference account: one Central Ohio territory, a named watchlist, source-backed change detection, Friday delivery and a learning loop that improves relevance from real operator feedback.
The working loop
Evidence → review → decision → outcome → better watchlist.
Signals can be marked useful, irrelevant, false positive, monitoring or actioned. Owners and deadlines are recorded. Approved outcomes become durable receipts and inform future briefs.
33Named watch items configured for the founding territory.
0Filler items required. If the evidence is not relevant enough, the publisher does not manufacture a brief.
100%Human approval required before a publisher output becomes an approved delivery.
Useful on day one
Ask the record, not the open web.
An MCP client can combine these questions with its own reasoning while RAN supplies the controlled evidence layer.
What changed in our monitored territory since the last approved Friday brief? Show the official sources and prior-record lineage.
Which open project findings need re-review because a source document changed? Include exact locators.
Show every Decision Receipt about material pricing, the owner, rationale, eventual outcome and cited evidence.
What has Division 7 consistently marked irrelevant, and which watchlist changes were human approved as a result?
List our provisional Roof IDs. For each one, show the source-backed facts and the unknowns that must be verified before it can be confirmed.
Search only the private inspection fields our owner approved. Show the source fingerprint, retention deadline and Roof ID or project link for every result.
Connect
One secure endpoint.
Use remote HTTP MCP in a compatible client. Create a separate token for each AI workspace so access can be expired or revoked independently.
https://www.roofingauthority.network/api/mcp
{
"mcpServers": {
"ran-intelligence": {
"type": "http",
"url": "https://www.roofingauthority.network/api/mcp",
"headers": {
"Authorization": "Bearer ran_mcp_YOUR_TOKEN"
}
}
}
}Configuration keys vary by client. Never paste a token into a public prompt, shared document or source repository. The token is shown once when created and stored hashed by RAN.
Current surface · v1.1.0
All 37 production tools.
The tool catalog spans nine evidence domains. Each response preserves the permissions, uncertainty and provenance of the underlying RAN Evidence Graph.
Project evidence
list_projects · get_project
search_project_documents
get_document_revision · list_findings
get_finding_evidence · compare_revisions
Industry record
list_episodes · search_episode_transcripts
get_transcript_passage · trace_quote
ask_network · get_topic_dossier
compare_speaker_perspectives
Market context
search_industry_signals
get_signal_sources
get_signal_context
Operator learning
list_intelligence_decisions
get_intelligence_decision_receipt
get_intelligence_learning_profile
list_intelligence_learning_changes
Change Publisher v2
list_intelligence_changes
get_intelligence_change
get_intelligence_publisher_output
Roof ID v1
list_roofs · get_roof
get_roof_evidence
list_roof_unknowns
Private context v1
list_private_context_sources
search_private_context_records
get_private_context_record
list_private_context_access_events
Bid Truth v1
list_bid_truth_opportunities
get_bid_truth_opportunity · get_bid_truth_assessment
list_bid_truth_evidence_gaps
get_bid_truth_decision_receipt
Paid boundary
pilot · trialing · active
Requires mcp_read_only
Inactive subscriptions are rejected.
Trust contract
Useful access. Narrow authority.
MCP is the paid delivery surface over the RAN Evidence Graph. Official sources remain official. RAN sells the work of finding, normalizing, versioning, ranking and connecting that evidence to an operator’s real decisions.
Tenant + grant bound
Private results require the token’s organization, an approved field-level grant and—when private context is requested—an opted-in token scope.
Source attached
Document locators, transcript timestamps, publisher URLs and evidence hashes travel with results.
Read only
The agent cannot approve, edit or delete projects, grants, feedback, decisions, policies or publisher records.
Revocable
Tokens are hashed at rest, expire automatically and can be revoked independently.
Questions
What paid access means.
The distinction matters: RAN does not charge for ownership of public facts. It charges for an accountable intelligence system built around them.
Is the MCP public?
The endpoint and tool schemas are discoverable. Every private evidence call requires a valid tenant-bound token and an active paid entitlement.
What are we paying for?
Normalized and versioned evidence, operator-specific relevance, decision history, managed delivery and a secure agent-access layer. Customers are not charged for ownership of their source-system data.
Can the agent take action?
No. The MCP retrieves evidence only. Approval, ownership, deadlines, outcomes, access grants and publisher decisions remain inside the Desk.
Can private system data appear automatically?
No. The source, purpose, objects, fields, persistence, retention, derived use and MCP rights must be approved first. A separately opted-in token is also required.
RAN Intelligence MCP
Make the agent show its work.
Start with a managed Founding Operator pilot, then connect the evidence to the AI workspace your team already uses.