NewTesela AI is here — AI agents for library design, protocols & optimization. Try it free→

MCP Server Privacy

This section supplements, and is governed by, the TeselaGen Privacy Policy.

Model Context Protocol (MCP) Server

TeselaGen offers a Model Context Protocol (MCP) server that lets third-party AI assistants and connectors (for example, Anthropic Claude and OpenAI ChatGPT) access your TeselaGen account on your behalf. This section describes how that integration handles your data. It supplements, and is governed by, the rest of this Privacy Policy.

What the MCP server accesses

When you connect an AI assistant to TeselaGen through the MCP server and ask it to perform a task, the server executes that task against:

  • Your TeselaGen LIMS account — for example, searching and retrieving DNA and protein sequences, reading and creating Electronic Lab Notebook entries, annotating sequence parts, creating chemical-compound and molecular-structure records, and submitting design/assembly jobs. The server acts only within the workspace(s) your account is permitted to access.
  • Public scientific databases — for tools you invoke that query external sources such as PubMed/NCBI, UniProt, PubChem, KEGG, GenBank, the RCSB PDB, and protein-analysis services. Query terms you provide are sent to those third parties under their own privacy policies.

The MCP server brokers these requests; it does not persistently store the content of your LIMS records. Data you request is returned to your connected AI assistant for the duration of the request. The server logs operational metadata only: each tool invocation emits one audit record containing the tool name, the authenticated user and tenant identifiers, outcome, and duration — never the tool arguments, the query content, or the response payload. Error records carry the exception type, not user data. Application logs carry a per-request correlation ID for tracing; request and response bodies are not logged.

How your data flows

  1. Your AI assistant sends a tool request to the MCP server over an authenticated, encrypted (HTTPS/TLS) connection.
  2. The MCP server forwards the request to TeselaGen's processing layer and, where applicable, to your TeselaGen LIMS instance, using your authorized credentials.
  3. Results are returned to the MCP server and then to your AI assistant.
  4. For tools that query public databases, the relevant query is sent to that external service.

Your AI-assistant provider (e.g. Anthropic, OpenAI) processes the conversation and tool results under its own privacy policy and terms; TeselaGen does not control that provider's handling of data once it is returned to your assistant.

Authentication and tokens

  • Access uses OAuth 2.1 with PKCE. You authenticate to your TeselaGen instance and grant the connector access; the connector never receives your TeselaGen password.
  • The MCP server issues a bearer access token scoped to your account and to a single permission scope (mcp). Access is further limited by your account tier.
  • The underlying TeselaGen LIMS credential is encrypted at rest (Fernet) and stored in the server's session store (Redis). Tokens are namespaced per tenant/deployment so one deployment cannot read another's state.
  • Tokens are time-limited: access tokens expire after a short interval (default 1 hour) and refresh tokens after a fixed period (30 days), after which re-authorization is required.
  • You can revoke access at any time. To revoke access, disconnect the TeselaGen connector in your AI client (Claude or ChatGPT), which discards the authorization grant; access tokens expire after 1 hour and refresh tokens expire after 30 days. You may also revoke the (Settings → API Password), which immediately terminates the server's ability to act on your behalf. The server automatically revokes an entire token family if refresh-token replay is detected.

Data retention

The MCP server retains OAuth/session state (encrypted credentials, token records) only for the lifetime of the tokens described above; expired records are removed automatically. Operational logs are stored in Google Cloud Logging (the _Default bucket) and retained for 30 days, after which they are automatically deleted; no logs are exported to external storage. Google Cloud admin-activity audit logs are retained separately for 400 days, as mandated by Google Cloud, and are immutable. TeselaGen uses no third-party analytics or telemetry for the MCP server.

Regulated / sensitive data

No tool is designed to store or return protected health information (PHI). Two tools, get_eln_entries and create_eln_entry, read and write free-text Electronic Lab Notebook content supplied by the customer. This free-text field could, at the customer's discretion, contain health-related information. TeselaGen does not intend the connector for PHI and instructs customers not to enter PHI into these fields. Customers operating regulated or clinical workloads should engage TeselaGen for a Business Associate Agreement before using ELN features.

Your AI-assistant provider

Because an MCP connector routes your data through a third-party AI assistant, that provider's privacy policy also applies to data you send and receive through the connector. Review your provider's policy (e.g. Anthropic, OpenAI) for how it handles conversation and tool data.

Contact

Questions about this section or how the MCP server handles your data can be directed to contact@teselagen.com.