Credal
Credal is an enterprise platform for building and governing MCP servers and AI agents. It connects over 1,000 data sources, inherits permissions automatically, and publishes governed agents to Claude, ChatGPT, Cursor or Slack.
What is Credal?
Credal is an enterprise platform for building, deploying and governing MCP (Model Context Protocol) servers and the AI agents that use them. It sits as a gateway between an organisation's agents and its systems, so authentication, permissioning and auditing are built once and applied everywhere instead of being reinvented by every team.
Three products make up the line. The MCP Platform connects more than 1,000 data sources — Google Drive, Slack, Confluence, Salesforce, Snowflake, SharePoint, Jira and Zendesk among them — and lets domain experts assemble each into a focused, department-specific server. Agent Builder offers two routes to the same governed agent: a no-code canvas of instruction, knowledge, tool and guardrail blocks for business experts, and an API and SDK for engineers who prefer to version, diff and test agents like code. Agent Registry keeps one inventory of everything deployed, with verified or draft status, a named owner, an approval chain and rollback.
The thesis running through all of it is that over-provisioned agents are expensive and inaccurate for the same reason: too much context. Credal therefore scopes each server down to the tools an agent genuinely needs, enforces standing instructions and pins call parameters deterministically. The homepage claims an 87% cut in context per query and a 73% gain in answer accuracy; the Checkr case study reports Salesforce query accuracy rising from 13% to 86%.
Permissions are inherited from source systems and enforced at query time rather than only at ingestion, so a user who cannot open a document in Google Drive will not see it through Credal either. A single consolidated audit log records every tool call and doubles as a cost ledger, attributing spend by team, workflow and model and flagging where a cheaper model holds accuracy.
Credal is model-agnostic — Claude, ChatGPT, Gemini, self-hosted open-source models or a customer's own Azure OpenAI deployment — and ships as multi-tenant cloud, managed single tenant, cloud-prem inside the customer's AWS account, or air-gapped on-premises. Named customers include Wise, MongoDB, Comcast NBCUniversal, Lattice, Checkr, incident.io and the U.S. Department of Health and Human Services.
What it does
- Turn any connected system into a governed MCP server
- Scope each agent down to the handful of tools it actually needs
- Enforce standing instructions and hard parameter limits on every tool call
- Inherit and enforce source-system permissions at query time
- Publish the same governed agent to Claude, ChatGPT, Cursor, Slack or your own API
- Require human approval before a sensitive action is allowed to run
- Attribute every dollar of AI spend to a team, a workflow and a model
When to use Credal / When not to
A quick filter to help you decide if Credal is the right fit.
When to use Credal
- Platform and IT teams opening up AI without rebuilding auth, permissioning and auditing for every project
- Security, GRC and compliance leads in regulated sectors who need SOC 2, HIPAA and GDPR controls on every tool call
- RevOps, support, engineering, HR and legal teams that want a governed MCP server dedicated to their own domain
- Organisations with strict data residency requirements, from managed single tenant to air-gapped on-premises
- Companies of more than 50 employees, the threshold Credal itself sets for its free trial
When not to use Credal
- Individuals and small businesses: there is no permanent free plan and no published price
- Teams under 50 employees, who are excluded from the free 14-day trial
- Anyone wanting a self-serve signup, since every path on the site leads to a sales demo
- Mobile-first users, as Credal ships no iOS or Android application
- Non-English-speaking teams, because the product, the site and the documentation are English only
How to use Credal
A typical end-to-end flow, from setup to results.
- Request a demo through the form at credal.ai/get-started, or email sales@credal.ai — there is no self-serve signup
- If your organisation has more than 50 employees, ask for the free 14-day trial offered on that same page
- Agree the deployment mode: multi-tenant cloud, managed single tenant, cloud-prem in your own AWS account or Kubernetes cluster, or air-gapped on-premises
- Sign in at app.credal.ai and connect your data sources from the admin Data Sources panel; permissions sync automatically
- Compose an MCP server by selecting only the tools a given team actually needs from each connected system
- Add enforced instructions, constrained parameters and guardrails, including human approval on sensitive actions
- Test the agent in the live sandbox before publishing anything
- Publish to the Agent Registry through its approval chain, then share it by team or role using RBAC
- Deploy the same agent to Claude, ChatGPT, Cursor, Slack or your own application through the API
- Monitor usage, permissions and spend in the consolidated audit log, and export it to your SIEM
Pros & Cons
Pros
- Permissions inherited from source systems and enforced at query time, not merely at ingestion
- Governance, auditing and legal review done once and applied to every team and every surface
- Real cost visibility: spend attributed by team, workflow and model, with quantified optimisation advice
- Model-agnostic with no vendor lock-in, including bring-your-own-key and in-VPC model access
- Four deployment modes, up to air-gapped on-premises, for strict data residency requirements
- SOC 2 Type 2, HIPAA, CCPA and EU-U.S., UK and Swiss Data Privacy Framework certification
- Written commitment not to train models on customer data, plus zero-data-retention agreements with providers
Cons
- No public pricing at all: a single Enterprise plan, quoted case by case
- No permanent free plan, and the 14-day trial is restricted to organisations of more than 50 employees
- No self-serve path — a sales demo is the only way in
- No mobile application, and product, site and documentation are English only
- The subprocessor list is not openly published: it comes via the trust portal or on request to security@credal.ai
- No postal address anywhere on the site, no Article 27 EU representative and no named DPO
- The Limited Use Policy, which carries the no-training commitment, has not been updated since September 2023
Pricing & Plans
Credal publishes no prices. A single Enterprise plan is quoted case by case and priced around four components: builder seats, user seats, data indexing volume and usage-based model tokens. There is no permanent free plan and therefore no entry-level price point in any currency. A free 14-day trial is available, but only to organisations with more than 50 employees, and a time-boxed paid pilot is the usual point of entry.
- Enterprise — custom pricing scoped around your teams
- data and usage
- billed on builder seats
- user seats
- data indexing and usage-based tokens
- Free 14-day trial — no charge
- restricted to organisations with more than 50 employees
Data, GDPR & hosting
A consolidated view of how Credal handles your data.
GDPR overview
GDPR compliance is claimed explicitly and backed by concrete mentions. A GDPR badge sits beside SOC 2 and CCPA on the homepage, and the pricing page lists SOC 2 Type II, GDPR and CCPA as included in every plan. The privacy policy, last updated on 8 June 2026, devotes a section to EEA and UK individuals: lawful bases (contract, legitimate interest, legal obligation, withdrawable consent) and rights of access, rectification, erasure, restriction, portability and objection, plus the right to complain to a supervisory authority. Credal AI is certified under the EU-U.S. Data Privacy Framework, its UK Extension and the Swiss-U.S. DPF, with free recourse through JAMS. A Data Protection Addendum template is published. Two gaps remain: no Article 27 EU representative and no named DPO. Privacy contact: privacy@credal.ai.
Who owns the data?
Credal's Limited Use Policy is unusually direct: the data you sync stays yours. Its use is limited to delivering Credal functionality, only Credal support staff can reach it, and even then only with your explicit permission. No advertising platform, data broker or information reseller is ever given access, and the data is never used to assess credit-worthiness or for lending. For customer data Credal acts as a service provider, so its own privacy policy does not govern that role, and employees of a customer must ask their employer about their rights. Credal does reserve the right to use aggregated, anonymised data as it sees fit.
Reuse rights
Credal states plainly that it does not train ML models on customer data, and it holds zero-data-retention or equivalent agreements with model providers including OpenAI, Anthropic and Cohere. Within that frame customers reuse their own synced data and the answers their agents produce without asking permission, and Credal actively encourages it: every AI interaction is logged so the organisation can build a proprietary record of its own AI traffic for later fine-tuning or model training. One asymmetry deserves attention. The full SaaS Agreement is not published and is only available on request, so the binding licence terms cannot be read before contacting the company.
Data retention & training
Hosting summary
Credal names Amazon Web Services and Microsoft Azure as infrastructure subprocessors, and says its AWS configuration is managed by an outside specialist consultancy. Four hosting arrangements are offered: standard multi-tenant cloud, a managed single-tenant instance with ingress and egress controls, cloud-prem inside the customer's own AWS account or Kubernetes cluster, and air-gapped on-premises. The stated principle is to keep data in-VPC wherever possible, so that data never has to leave the customer's environment. Jurisdiction is the United States: the privacy policy warns that information may be transferred to other countries including the United States, and the Data Privacy Framework section confirms Credal holds personal information there. No European hosting region is advertised and no country other than the United States is named. The domain resolves to a Google anycast node, which reflects the CDN in front of the marketing site and says nothing about where customer data is actually stored.
Things to keep in mind
Risks and trade-offs to weigh before adopting Credal.
- Making company knowledge instantly queryable can quietly retire the habit of checking answers at source; Credal traces answers back to the originating document, but the discipline still has to be maintained
- Permissions are only as good as the systems they mirror: a badly permissioned Google Drive or Salesforce becomes a badly permissioned agent
- Usage-based token billing means cost scales with adoption, so a successful rollout can produce an unexpected bill
- The headline figures — 87% less context, 73% more accuracy, 13% to 86% at Checkr — are vendor-reported and not independently audited
- Routing every tool call through one gateway concentrates risk: an outage or a misconfiguration there reaches every team at once
- The audit log is a detailed record of who asked what, which is valuable for governance and equally a surveillance surface if misused internally
- The full SaaS Agreement is not published and only available on request, so binding terms cannot be reviewed before engaging with sales
Setup & Integrations
Technical difficulty
Moderate, and front-loaded. The no-code canvas is genuinely aimed at non-technical domain experts, but the initial setup is an IT project: connecting sources, wiring SSO, SAML and SCIM, and choosing a deployment mode. Credal says most teams are running within days, with sources connected and a first workflow built in one session, and a full rollout taking a few weeks. Cloud-prem and air-gapped options require in-house AWS or Kubernetes skills. The OpenAI/Anthropic-compatible gateway is the easiest piece: a one-line configuration change, no code modification.
Deployment
Integrations
Supported languages
Behind Credal
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Is there a free trial?
How much does Credal cost?
What deployment options are available?
Does Credal train AI models on customer data?
Which security certifications does Credal hold?
Can we bring our own AI models?
How are data permissions handled?
How long does implementation take?
Is there a mobile app?
Is there a minimum age?
Should you pick Credal?
Credal is a serious, narrowly aimed product: a governance layer sitting between an organisation's AI agents and the systems those agents touch. Its central claim — that over-provisioned agents are expensive and inaccurate for the same reason — is a genuine insight, and the platform is built around it consistently, from scoped tool sets and enforced parameters to a cost ledger attributing every dollar of model spend to a team and a workflow.
The security posture is unusually thorough for a company of this size. SOC 2 Type 2, HIPAA configurations, Data Privacy Framework certification, zero-data-retention agreements with model providers, air-gapped deployment and a written commitment not to train on customer data all point the same way. The customer list — Wise, MongoDB, Comcast NBCUniversal, Checkr, the U.S. Department of Health and Human Services — is checkable and consistent with that positioning.
The reservations are commercial rather than technical. Nothing about the price is public: one Enterprise plan, four billing components and a mandatory sales conversation. The free trial exists but is closed to organisations under 50 employees, which is an honest signal about who this is for. A few transparency gaps sit oddly beside the security effort: no postal address anywhere on the site, no Article 27 EU representative despite EEA ambitions, a subprocessor list reachable only through the trust portal or on request, and a Limited Use Policy — the document carrying the no-training promise — untouched since September 2023.
For a mid-sized or large organisation already running AI across Claude, ChatGPT, Cursor and Slack, and already worried about who can see what, Credal answers a real and awkward question. For anyone smaller, or anyone who needs a price before talking to a salesperson, it is simply not addressed to them.
- Choosing a selection results in a full page refresh.
- Opens in a new window.