MCP connector storage and logging
Updated 6 November 2026
/agent-integrations/mcp-connector/security/storage-and-logging
MCP connector storage and logging
TechWolf keeps a record of each call, with its inputs but never the data it returned. The whole record, inputs included, is kept for 400 days in your region.
What TechWolf stores
| Record | What it holds | Personal data | Where | Kept for |
|---|---|---|---|---|
| Call records | Who called, when, from which AI tool, which tool, its inputs (with the reason), its status (success, denied or failed) and the duration | The caller's user ID; names a person typed into an input | Your region | 400 days, inputs included |
| AI tool registrations | The name an AI tool registered under and its sign-in return address | No | Your region | 365 days after last use |
| Sign-ins in progress | The state of a sign-in that has not finished | No | Your region | 10 minutes |
| One-time sign-in codes | The organisation, user ID and email of the person signing in | Yes | Your region | 5 minutes, or until used |
| Rate-limit counters | The organisation ID and a one-way hash of the person's ID | Pseudonymised | Your region | One value per person, overwritten on each call |
| Lists held in memory | The employee, job and job family lists fetched for one person | Yes | Your region, memory only | Up to 5 minutes, for that person only |
| Service traces and logs | Timing, the tool, the reason, inputs with free text reduced to its length, a one-way hash of the person's ID, the AI tool's network address | Pseudonymised | Your region | Traces 30 days, logs 400 days |
Returned data is never stored. It passes through the connector's memory on its way back to the AI tool. The only exception is the short-lived lists above, which save the person waiting on repeated lookups.
Employees in results appear under the identifiers your organisation loaded into TechWolf, with names where you loaded them.
No conversation memory. The connector keeps no chat history and no memory between calls. Your AI tool keeps the conversation, under its own settings.
Deletion. On request, or at the end of your contract, TechWolf deletes your organisation's call records. Send requests to privacy@techwolf.ai.
What a call record holds
Every call by a signed-in person is recorded with:
- The time of the call and its duration.
- The organisation ID and the caller's user ID.
- The tool called and its status: success, denied or failed.
- The AI tool, as it identifies itself, its MCP session ID and protocol version.
- The inputs, including the one-sentence reason.
The reason field. Every tool requires a short reason, which the AI tool writes. It is usually the person's question in one sentence. The AI tool is instructed never to put personal or other sensitive details in it. The reason lets your organisation and TechWolf see why the data was read.
Rebuilding what happened. TechWolf holds no prompt and no model reasoning. To trace a conversation end to end, combine your AI tool's own logs with TechWolf's call records, matched on the MCP session ID and the time.
Logs, traces and error reports
Service traces and logs support operations. They keep the reason, but reduce other free-text inputs to their length and replace the person's ID with a one-way hash. They keep only the size of a result, never its contents, and they stay in your region.
Error reports are sent only when a call fails unexpectedly, never for a denied call or invalid input. They hold the call's inputs, with emails, user IDs and access tokens removed, and never the data the call returned.
Sign-in secrets, such as one-time codes and tokens, never appear in logs or traces.
Access and audit
Who at TechWolf can read them. Authorised TechWolf engineering and support staff, to run and support the connector. Staff access is role-based, uses SSO with MFA and is logged.
Audit for your organisation. On request, TechWolf provides an extract of your organisation's call records, for example for an investigation. There is no self-service log export.
Monitoring. Errors, latency and availability are monitored and raise alerts to TechWolf's engineering team.