Your AI assistant · MCP
Use Dashport from your AI assistant.With the numbers you approved.
Ask the assistant you already use about sales, stock or margin, and it answers from Dashport: only definitions your team approved, only data your permissions allow, and every call written to your audit log. Any MCP client can connect.
- The numbers you approved
- With your permissions
- Every call logged
Your AI assistant, connected to Dashport
Try a question
You: What was Q1 revenue by customer segment?
Assistant · answered from Dashport
Enterprise led Q1 2026 with $62,570, then Mid-market with $55,210 and Small business with $17,770.
What Dashport checked
- Revenue uses the definition you approved (v4)
- 9 rows in Q1 · nothing missing
- Only the Northstar source in this grant
- Logged · trace 3b9e
Underneath · the MCP calls
semantics_search("revenue")
✓ Allowed→ 3 published concepts
query_plan("Q1 revenue by segment")
✓ Allowed→ SUM(revenue) by customer_segment · Q1 2026
query_execute(plan)
✓ Allowed→ 3 rows · complete · receipt attached
Every call, allowed or blocked, lands in your audit log.
Assistant + Dashport
Your AI assistant can answer a question.
Dashport keeps your whole business on one board: connected once, checked, refreshed and shared with the right people.
Full comparisonHow it works
Three promises, kept by Dashport, not by the assistant.
Your assistant is good at conversation. The rules live in Dashport, so they hold the same way whichever assistant asks, and the same way they hold for your team.
- 01
The numbers you approved
Your assistant searches the definitions your team approved and plans from them. A metric that isn't approved gets a plain no, never a guess.
semantics_search · semantics_get · query_plan
- 02
With your permissions
It signs in as you, for one workspace, with the sources you choose. Masked columns stay masked. Revoke it whenever you like.
OAuth + PKCE · one workspace per grant
- 03
Every call logged
Every call, allowed or blocked, lands in the governed agent timeline of your audit log, with what was asked and what came back.
governed agent timeline
Connect your assistant · MCP
Connect any MCP client. No passwords in the file.
Add the Dashport gateway to your assistant. On first connect you sign in with your own account, choose the workspace and the sources, and your assistant works through the same rules as everyone else. Revoke it whenever you like.
https://v2agents.dashport.ai/mcp- OAuth with PKCE
- Dynamic client registration
- Rotating refresh tokens
- One workspace per grant
- Revocable any time
Your MCP client
any client
Dashport sign-in
v2auth.dashport.ai
MCP gateway
v2agents.dashport.ai
1 · Register
dynamic client registration
2 · Authorize with PKCE
you sign in and consent
3 · Tokens
access + rotating refresh
4 · Call a tool
checked against your permissions
5 · Governed result
with its receipt · logged
The handshake
No keys to paste, no database password to share. Your client registers itself, you approve it once, and every call after that is checked.
Consent, in plain language
The consent screen names the workspace, explains each scope in a sentence and lets you narrow the sources. Try it.
Agent authorization
Your MCP client
This client is asking to use Dashport on your behalf. It never receives a database password, and every request stays limited by your workspace permissions.
Workspace
Northstar· Team workspace
This authorization can't switch to or read another workspace.
Requested access
- Discover semantic concepts
- Read the semantic model
- Plan governed questions
- Execute governed queries
- Read governed results
- Read agent conversations
- Create and continue conversations
What each scope means
- Discover semantic concepts
- Search the published model for governed business concepts and metrics.
- Read the semantic model
- Read approved concept, metric, relationship and classification details.
- Plan governed questions
- Create explainable query plans within your authorized data scope.
- Execute governed queries
- Run approved plans with policies and result transformations enforced.
- Read governed results
- Receive policy-approved results and their verification receipt.
- Read agent conversations
- Read only conversations created by this agent in this workspace.
- Create and continue conversations
- Create persistent, governed multi-turn analytics conversations.
Dataset permissions, row and column policies and result checks still apply. You can revoke this grant later in Agents & MCP.
Interactive mock
Your assistant asks. Dashport decides.
Every call is planned against the published model, checked against your permissions and written to the governed agent timeline, allowed or not.
Audit log · governed agent timeline
trace 3b9eYou → Your MCP client · Northstar
- ✓ Allowed
semantics_search("revenue")
→ 3 published concepts
- ✓ Allowed
semantics_get("gross_margin")
→ v4 · weighted by revenue
- ✓ Allowed
query_plan("Q1 revenue by segment")
→ SUM(revenue) by customer_segment · Q1 2026
- ✓ Allowed
query_execute(plan)
→ 3 rows · complete · receipt attached
Enterprise $62,570 · Mid-market $55,210 · Small business $17,770
- ✕ Blocked
semantics_get("net_revenue_retention")
→ not a published metric
- ✕ Blocked
query_execute(source: Stripe)
→ source not in this grant
Seven governed tools
Small, predictable tools. Each one runs inside the grant and leaves a receipt.
| Tool | What it does | What always holds |
|---|---|---|
| semantics_search | Searches the published model for metrics and concepts | ✓ published concepts only |
| semantics_get | Reads one concept: definition, relationships, classification | ✓ approved details only |
| query_plan | Turns a question into an explainable plan | ✓ inside your data scope |
| query_execute | Runs an approved plan and returns the result | ✓ policies and masking enforced |
| conversation_create | Starts a governed multi-turn conversation | ✓ one workspace per grant |
| conversation_get | Reads a conversation | ✓ only ones this agent created |
| conversation_continue | Asks a follow-up in the same conversation | ✓ same rules as the first turn |
Connect any MCP client
Add the gateway to your client's MCP config. On first connect it opens Dashport sign-in, so no password or API key goes in the file.
{ "mcpServers": { "dashport_northstar": { "type": "http", "url": "https://v2agents.dashport.ai/mcp" } }}Agents & MCP in the app prints this config for your workspace, plus a ready-made one for Hermes.
Inside Dashport
Inside Dashport, agents do the groundwork. You decide.
The same rules hold for Dashport's own agents. They answer questions, draft your business definitions, watch your boards and send the weekly report. Each one shows its work, and nothing becomes a number you rely on until you approve it.
One morning on the Northstar sample
- done
- needs you
- blocked
Agent
Scheduler · 08:00
Weekly review sent
sentWork log
The agents
Four kinds of agent. Each with a job, and lines it can't cross.
Turn a card over to see its work log and what it is not allowed to do. Those limits are enforced by the server, not left to the agent's judgement.
Chat agent
Answers questions
Plans the query, runs it, checks the result and draws the chart, then offers the next step. Pick a source, ask in plain language, open the SQL any time.
- plan
- query
- ✓ check
- chart
Follow-up tools
- export_csv
- export_excel
- generate_chart
- show_data
- drilldown
- compare_period
- segment_breakdown
- add_to_dashboard
- share
Chart types
line · bar · area · pie · treemap · KPI · table · list · cards · text · progress · gauge
Setup agents
Prepare your workspace
Draft the slow parts of BI setup from metadata. Every output is a proposal: nothing becomes truth until you accept it.
Proposals you review
- Semantic model draftfrom metadata only, no rows→ ✓ / ✕
- Relationship suggestionssuggested joins with confidence→ ✓ / ✕
- Classification scansensitive fields, suggested levels→ ✓ / ✕
- Sheet structuregenerate with AI, then review→ ✓ / ✕
Board analyst
Watches your boards
Reads a Superdashboard and writes AI Analysis: insights labelled trend, anomaly, suggestion or opportunity, each with its confidence.
AI Analysis
- Trend92%Revenue +21.5% Jan → Apr
- Anomaly88%Europe −35.8% in April
- Suggestion74%Add margin by segment to the board
- Opportunity61%Small business is 15.2% of revenue
Scheduler
Sends the report
Turns any chat answer into a scheduled report: daily, weekly or monthly, at the time you set, to workspace members.
Schedule
The principle
Agents suggest. Rules decide.
Models are good at drafting, planning and noticing. They are not a place to keep rules. So Dashport splits the work: agents assist and show their confidence; the server enforces, the same way every time.
~ Suggests
What agents do
Draft, plan, suggest, explain. Fast and useful, sometimes wrong, so every output is shown and most are proposals.
- Draft a business model≈ proposal
- Plan a query≈ shown in full
- Suggest a join≈ 0.94
- Flag an anomaly≈ 88%
= Decides
What the server enforces
Not a model's opinion. Checked on every request, for people and agents alike.
- Published metrics only✓ always
- Your permissions, never more✓ always
- Masked stays masked✓ always
- Complete data, or no chart✓ always
- Every call in the audit log✓ always
Ask from anywhere. Get the numbers you approved.
Start with the Northstar sample: connect your assistant, ask a question and find the call in your audit log. Then connect your own data.
14-day free trial · Cancel anytime · Secured by Stripe