MCP connector security
Updated 6 November 2026
/agent-integrations/mcp-connector/security
MCP connector security
The MCP connector lets your AI tool read your TechWolf skills and workforce data for the person using it. It only reads, it runs no AI model, and each person sees only what they can already see in the TechWolf Console.
How it works
Shaded areas are run by TechWolf. Market search is the only tool that reads outside your region.
- A person asks their AI tool a question, such as "Which skills does the data analyst job need?"
- The AI tool calls a TechWolf tool. The connector checks that person's access and reads only the data they may see.
- The AI tool writes the answer. The conversation itself stays with your AI tool and never reaches TechWolf.
Why market search can leave your region. TechWolf's index of public job postings is hosted in the US. If your TechWolf data is in the US, market search stays in your region. Elsewhere, only the search terms go to the US, with no identity attached.
Key attributes
Read-only Every tool only reads. Nothing can be created, changed, deleted or sent. |
No AI model at TechWolf The model is your AI tool's, so nothing that passes through the connector trains a TechWolf model. |
The person's own access Every call checks the person's TechWolf Agent role and their Console data access. |
What TechWolf sees Only the tool call: the tool, its inputs and a one-sentence reason. Never the chat, the full prompt or the answer. |
What TechWolf keeps A record of each call, never its results. The whole record, inputs included, is kept for 400 days. |
Your region The connector runs on AWS in the region that holds your TechWolf data: EU, US or Canada. Market searches go to TechWolf's index of public job postings in the US, without any identity. |
Your sign-in People sign in with your organisation's SSO. The AI tool gets a one-hour token for that person, never a password. |
Switching it off Remove a person's TechWolf Agent role, or ask TechWolf to switch the connector off for your whole organisation. |
Assurance
TechWolf holds ISO/IEC 27001 and ISO/IEC 42001 certifications and SOC 2 Type 1 and Type 2 reports, available on request at trust.techwolf.ai. Security questions: security@techwolf.ai. Privacy questions: privacy@techwolf.ai.
Go deeper
Each page below answers one part of a security review in full.
- Data flow and hosting. What reaches TechWolf in a tool call, where the connector runs and what leaves your region: MCP connector data flow and hosting.
- Storage and logging. Every record TechWolf keeps, what it holds, for how long and who can read it: MCP connector storage and logging.
- Sign-in and access control. OAuth sign-in, tokens, the checks on every call and tenant isolation: MCP connector sign-in and access control.
- AI and threat model. The AI model, training, and how prompt injection and other AI threats are contained: MCP connector AI and threat model.
- Controls and shared responsibility. What your admins control, the rate limits, and who secures what: MCP connector controls and shared responsibility.
- Operations and assurance. Availability, change management and TechWolf's security programme: MCP connector operations and assurance.
- Security FAQ. Short answers to the questions security reviews ask most: MCP connector security FAQ.
- Tools. Each tool, what it reads and the data access it needs: MCP connector tools.
- Access. How the TechWolf Console admin gives people the TechWolf Agent role and their data access: Give people access to the Analyst Agent and the MCP connector, which opens after sign-in.