Veriom MCP
Your assistant works in Veriom. Decisions stay with you.
Connect Claude, Cursor, VS Code, ChatGPT, Codex, Windsurf or Gemini CLI. It reads your findings, audits and architecture, runs audits and triages findings as you, within your role and the scopes you allow.

Which critical findings are open in payments-api?
One: the ledger query is built from request parameters, in src/ledger/queries.py:88. I can ask Veriom to prepare a fix for you to approve.
- veriom_list_findings
- veriom_get_finding
- read-only
- Signed in as you
- findings:read
- Approving a fix stays with you
Integrations
Your sources in. Your assistants on top.
Sources send Veriom evidence, never instructions, and every finding cites where it came from. Assistants then work on what Veriom found, as the person who connected them. Hover, tab to or select a logo to follow its path.
- Private access
- GitHub, GitLab, Bitbucket, GitHub Actions, AWS, Azure, Google Cloud
- In validation
- Google Drive, Notion, Confluence, Slack, Discord, Jira, Linear
- MCP clients
- Coding and chat assistants read findings, run audits and triage findings as the person who connected them. Workspace agents are in development.
Context flow
See a source become reviewable architecture context.
Choose a source family to see the observation Veriom keeps, what corroborates it, and the relationship a person or a bounded agent can inspect next.
- 01ObservedA route registration and a deployment binding describe the same service path.
- 02CorroboratedBoth name payments-api, and the deployment ran from the same commit.
- 03ReconciledRelease pipeline deploys payments-api
Illustrative. The knowledge and team-operations connectors shown here are still in validation.
From connection to model
Four gates before evidence supports a claim.
What an assistant sends is authorised, validated and reconciled before the model uses it, and what it reads back cites the evidence it rests on. It never reaches your sources directly.
- 01AuthoriseYour sign-in or API key sets the workspace, your role and the scopes the client may use.
- 02ValidatePayload, archive, size and source rules are checked before ingestion.
- 03ReconcileAgreement, contradiction, freshness and superseded facts are resolved.
- 04CiteFindings and Ask Veriom answers point back to evidence IDs, in the app and over MCP.
Boundaries
Access as you, never beyond you.
Who a client acts as, what it can't decide, and how it treats what it reads.
It acts as you
A client signs in as a member of one workspace and works within your role and the scopes you allow.
Decisions stay with people
Approving fixes, reports and decisions, accepting a risk and merging stay in the Veriom app. No MCP grant can carry an approval.
Findings are data
Finding titles, descriptions and evidence come from your code. The assistant reads them as data, never as instructions.
Tools say what they do
Read tools are marked read-only and the ones that delete are marked destructive, so your client knows when to ask you first.
What is available today.
- Read findings, audits, architecture, compliance programs and reports over MCPLive
- Run audits, triage findings and request fixes over MCPLive
- Approvals, accepted risk, merges, members and billing over MCPStay in the app
- GitHub, GitLab and Bitbucket repositories; AWS, Azure and Google Cloud collectorsPrivate access
- Slack, Discord, Drive, Notion, Confluence, Jira and LinearIn validation
- MCP-aware workspace agentsIn development
Questions teams ask
Clear limits before a clever workflow.
What can my assistant do through Veriom MCP?
Read the workspace, its repositories, audits, findings, architecture map, compliance programs and reports. Start, retry and cancel audits. Comment on, assign and prioritise findings and move them, for example to fixed. Ask Veriom to prepare a fix, and change audit schedules, triggers and webhooks. It does all of this as you, within your role and the scopes you allow.
What stays with a person?
Approving remediations, reports and decisions, accepting a risk, merging or deploying, autopilot and feature settings, members and sign-in, API tokens, billing and deleting a workspace. The tools refuse these, and no MCP grant can carry an approval scope.
Which clients can connect?
Claude Code, Claude Desktop and claude.ai, Cursor, VS Code, ChatGPT developer mode, Codex, Windsurf and Gemini CLI. Each signs you in through Veriom in your browser, or uses a workspace API key.
Does the assistant follow instructions found in my code?
It shouldn't. Finding titles, descriptions and evidence come from your code and from scanners, and Veriom tells the assistant to treat them as data, not as instructions. Secret values in evidence are masked before they reach it.
Are Slack, Discord, Google Drive and the other connectors live?
Not yet: Slack, Discord, Drive, Notion, Confluence, Jira and Linear are still being validated. Read-only GitHub, GitLab and Bitbucket repositories and bounded AWS, Azure and Google Cloud collectors are available in private access beside the MCP evidence gateway.
Ask your assistant what to fix first.
Add Veriom to your MCP client, sign in, and let it read your open findings and the evidence behind them.