Middesk
Also known as: Middesk Agents, Middesk Verify
Middesk helps banks, lenders and payment companies verify businesses and investigate issues before onboarding them. Its agents draw on corporate records, watchlists, web evidence and other sources to work through flags such as a possible sanctions match or an inconsistent address. The platform can apply findings under the account’s rules or hold them for an analyst. Each run keeps its findings, sources and resulting actions together.
Middesk verifies businesses and helps financial institutions manage risk from onboarding through monitoring. When a verification report raises a question, a buyer can start a Middesk agent from the Dashboard, an API or a policy. Its orchestrators handle customer identification, customer due diligence and enhanced due diligence.
An orchestrator can delegate a question about a business address or sanctions match to a specialist, collect the findings and propose what should happen next. Agents consult more than 400 government and authoritative sources, including registries, watchlists and web evidence, and tie conclusions back to the data used in the review. Their work covers sanctions false positives, address discrepancies, ownership verification, litigation research, merchant risk and operating status.
A run can pause with a set of findings, material from the relevant sources and confidence scores. An analyst can accept a finding, correct it or set it aside. The agent then applies the resolution to the Middesk report. Account administrators can keep every finding under manual review or set a confidence threshold for automatic application. The report retains a record of the decision and any actions the agent took. Webhooks can notify a connected application when an agent pauses, finishes or updates the business report. A business thread holds context and history across runs so a team can continue a review in the same place.
Banks, lenders, payment processors and companies with embedded financial services can use Middesk to verify identity before opening an account or activating a merchant. Middesk verifies 7 million businesses a year, with coverage in more than 65 countries, and its customers include Toast, Mercury, Brex, Ramp, Plaid, Rippling, Cash App, Shopify, Intuit, Gusto, Revolut and Marqeta. Rho went from 300 to 400 applications a month with five analysts to more than 2,000, with address validation accuracy rising from 70% to nearly 100%.
Novo integrated Middesk Verify and handled more than 10,000 applications per month with over 50% approved automatically. Checkr uses Middesk for customer verification within its background screening business. Middesk also supports monitoring after onboarding and entity services such as registrations and lien filings. A payments team can direct an agent to investigate a possible sanctions match, hold a disputed finding for review and use the final report in its onboarding process.
Middesk sells verification and agent access through its sales team. Its hosted application has separate production and sandbox environments for reviewing live applications and testing integrations with sample businesses. Developers can retrieve the steps and artifacts for an individual agent run or use the application’s APIs to respond to a review.
An MCP server lets an authorized assistant retrieve Middesk information, and a Claude connector runs business verification checks from inside Claude. The service undergoes annual SOC 2 Type II audits. Its Dashboard supports role based permissions, single sign on and directory sync, with account administrators deciding which staff can work in the live environment.
Vendor details
Canonical URL
https://www.middesk.com/
Category
Enterprise operations agent
Subcategory
Business verification and merchant due diligence agents
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
A REST API, GraphQL and webhook delivery connect onboarding applications to Middesk. The Middesk MCP server can expose supported business information to an outside assistant.
In practice
A payment platform routes a possible watchlist match to a specialist and asks an analyst to review the evidence before applying the result.
A lender starts an agent after verification flags an address and receives a report update once the findings are resolved.
An onboarding team sets a confidence threshold so routine findings apply automatically and uncertain findings wait for a person.
Sources & related URLs
Research sources
Agentic Index coverage score
9.5 / 14 capabilities · 68%
| Integrations & Tool Calling | Partial |
|---|---|
|
Agents can collect information from external business sources and link it to their findings. A report can gain additional sources, selected evidence or a corrected verification result after an agent completes a review. These actions change Middesk records. A buyer can receive a business update in its own system by subscribing to a webhook and fetching the revised report there. Sourcedocs.middesk.com/work-with-agents/take-actionread 2026-11-20 |
|
| Workflow Orchestration | Full |
|
Middesk orchestrators plan the work, dispatch specialist agents and consolidate the results. A customer identification review can delegate tasks involving a watchlist match, a business address or other components of a verification report. Each specialist returns material that the orchestrator can use to propose the next action. A run can then carry out a resolution when policy or an analyst approves it. Sourcedocs.middesk.com/work-with-agents/run-agentsread 2026-11-20 |
|
| Knowledge Grounding & RAG | Partial |
|
An investigation cross references registries, watchlists and other business sources with the details of the specific case. An agent can review a reported address or ownership flag, link the sources it used and create an explanation for the outcome. The report retains these selected references for the person reviewing the case and can receive additional source material after an investigation. Sourcemiddesk.com/solutions/use-cases/agentsread 2026-11-20 |
|
| Human Oversight & Guardrails | Full |
|
The interrupt policy pauses an agent when its findings need a decision. An account administrator can require manual review for every finding or set a confidence threshold for automatic application. Findings below the threshold still wait for review. A person can correct a finding or change its disposition and let the agent carry out the resulting action on the report. Sourcedocs.middesk.com/work-with-agents/interrupt-policyread 2026-11-20 |
|
| Security, Identity & Governance | Full |
|
Middesk undergoes an annual independent SOC 2 Type II audit for security, availability and confidentiality. Dashboard roles distinguish access to live reports, sandbox, orders and team administration. Eligible accounts can turn on single sign on, and Directory Sync can map identity provider groups to Middesk roles. Only admins can manage those account settings and assign team access. Sourcedocs.middesk.com/security/certifications-principlesread 2026-11-20 |
|
| Observability & Auditability | Full |
|
Each run retains steps and artifacts that show the work the agent performed. Steps can be nested to include delegation to a specialist, and an artifact records confidence and source references for a particular finding. A reviewer can see the resolution and changes to the report after application. The API can retrieve the relevant run for an outside audit workflow. Sourcedocs.middesk.com/work-with-agents/run-agentsread 2026-11-20 |
|
| Memory & State Persistence | Partial |
|
A Middesk thread keeps work context and history across multiple runs, usually associated with a business. The application can start another run in the same thread and request the accumulated execution steps. Threads have their own identifiers and store progress through completion, error or interruption. A later run can continue the case with that history available. Sourcedocs.middesk.com/reference/threadread 2026-11-20 |
|
| Deployment & Data Residency | Not documented |
|
Middesk runs its Dashboard and API as a hosted service, with separate production and sandbox environments. A production credential works against live requests, and a sandbox credential works against test orders. An account can simulate verification outcomes in the sandbox without changing live business records. The hosted environment and its test counterpart are operated by Middesk. Sourcedocs.middesk.com/environmentsread 2026-11-20 |
|
| Prebuilt Agents / Templates / Packs | Full |
|
An account can retrieve its available agents and select an identifier when it starts a run. The catalog has prebuilt orchestrators for customer identification and distinct specialists for tasks such as address verification or sanctions screening. The orchestration agent can delegate to a specialist when a case calls for it. The same buyer can invoke the agents on different businesses. Sourcedocs.middesk.com/reference/agentread 2026-11-20 |
|
| Triggers & Channel Coverage | Full |
|
A Middesk policy can start an agent when a business enters review after a verification order. Accounts with the customer identification, due diligence or enhanced due diligence orchestrator have matching default trigger policies. An operator can configure the policy to invoke the chosen agent. Manual starts from the Dashboard and programmatic starts through the API remain available. Sourcedocs.middesk.com/work-with-agents/run-agentsread 2026-11-20 |
|
| Model Flexibility & Routing | Not documented |
|
An agent run takes the agent identifier and context such as a business or a thread. Middesk supplies the agent definitions and the application selects among those available to the account. Buyers control the triggering conditions and review thresholds for investigations while the hosted service manages the reasoning components used to evaluate the merchant or business. Sourcedocs.middesk.com/api-reference/agents/runs/create-agent-runread 2026-11-20 |
|
| APIs / SDKs / MCP Extensibility | Full |
|
The Middesk API lets an authorized application create a run with an agent identifier and a business context, or place the run into an existing thread. API keys identify the requesting account. Application developers can retrieve the execution and use additional endpoints for agent threads and review actions. A connection can receive webhooks about results recorded in Middesk. Sourcedocs.middesk.com/api-reference/agents/runs/create-agent-runread 2026-11-20 |
|
| Testing, Debugging & Optimization | Partial |
|
The sandbox can create sample businesses and return a specified verification result without touching real accounts. A developer can test the way a connected application responds to sample identity outcomes and notifications before moving to the live environment. Within the agent product, the run record keeps steps and evidence that support diagnosing a particular finding. Sourcedocs.middesk.com/environmentsread 2026-11-20 |
|
| Browser / Computer-use | Partial |
|
An investigator can attach captured evidence from external web pages to the report. With screenshot capture enabled, the agent result includes the capture status and identifies which pages finished. When captures arrive after the business update, another notification indicates that the screenshots are ready. The buyer can retrieve the latest report after those captures finish. Sourcedocs.middesk.com/work-with-agents/review-workread 2026-11-20 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Quote for Middesk agent access
Contracted business identity products and agent access
What is public
Middesk makes agents available through its sales team and account managers. Its platform also offers verification, monitoring and registration products.
Billing mechanics
Purchased access determines which agents are available on the account. The Dashboard and API have a separate sandbox for testing.
Variable cost rationale
Agent availability depends on the account agreement. The price and any separate usage charges depend on the package purchased.
Sales call required
Yes, required for paid access
Commercial notes
Agent runs can begin from an eligible policy, the Dashboard or the API. Production access and test access use different credentials.
Missing data
Agent fees, included run allowances and usage rates are quoted.
Related vendors
- 11th Estate — Agentic platform whose AI engine scans markets, matches an…
- Abacum — FP&A platform for growing companies whose AI Modeler builds, edits…
- Accend — Accend builds human in the loop AI agents for commercial credit…
- Adopt AI — Finance agents that categorize, reconcile and close the books inside…
- Aera Technology — Decision intelligence platform where an always on agent executes…
- Akro AI — On premise operational intelligence that automates document heavy…
Alternatives to Middesk
Other products that do the work Middesk does, and what each one is.
- Cube
- Manhattan Associates
- Ocrolus
- RampFinancial operations platform
- Apprentice.ioAI-native manufacturing platform for regulated production
- CleavrAI collections coworker