Strand AI

Predict spatial proteomics from H&E slides with an AI agent.

Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.

Paste it into any MCP client. Setup docs

What the Strand AI connector does

The Strand AI MCP server connects Lattice, Strand AI's model that predicts protein-marker maps from routine H&E histology slides, to Claude and other MCP clients. Ten tools upload de-identified slides, price and submit prediction runs, segment cells and return results. Runs spend prepaid credits. Sign-in is OAuth, scoped to one Strand AI organization.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on .

Anthropic states this reflects the level of review a connector received, not a security audit.

Strand AI tools (10)

  • cancel_job
  • download_results
  • get_job
  • get_sample
  • list_markers
  • list_samples
  • segment_sample
  • submit_job
  • update_sample
  • upload_sample

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • Research use only. Strand states the service is not for clinical diagnosis, treatment or patient-care decisions. Lattice output is a prediction from H&E, not a measured stain.
  • Slides must be de-identified first. Strand's security page says not to upload protected health information unless a separate written agreement authorizes it. The requirement covers slide pixels, embedded label and macro images, scanner metadata and the filename.
  • Automated de-identification is off by default. When Strand enables it, it strips label and macro images and known metadata fields, but it does not redact identifiers in the tissue image or filename. Strand says it is "not a representation" that data meets HIPAA.
  • No self-serve credit top-up. Strand's pricing page says credits are sold in prepaid tiers and added by emailing support.
  • Spending guard is advisory on the server side. The estimate-then-confirm step is an instruction to the model. Your client's approval prompt is what actually stops a run.
  • An API key bypasses scope separation. A key has full authority across all ten tools, including spending credits.
  • One organization per connection. To switch organizations, Strand says you reconnect.
  • Uploads need a Claude egress change. Without storage.googleapis.com allowed, uploads from Claude fail. Large slides upload faster in the browser.
  • Tool detail not observed by us. The handshake was refused without credentials, so tool schemas and annotations on this page come from the directory and Strand's documentation.

Frequently asked questions

Does running Lattice from Claude cost credits?

Yes, a run spends your organization's prepaid credits, and nothing else does. Strand's docs say browsing, viewing, estimates and cell segmentation are free. One credit covers one retained tissue position on one marker, so a small biopsy costs less than a slide covered in tissue. Strand tells Claude to show the estimate and get your approval before starting a run.

What happens to my credits if a Lattice run fails or I cancel it?

They come back automatically. Strand's pricing page says a submitted run reserves its full estimated cost, and a failed or cancelled run is credited back in full. If a run delivers some markers and not others, you pay only for the delivered ones. A run your organization cannot afford is rejected outright rather than partially charged.

Why does uploading a slide from Claude fail with host_not_allowed?

Because Claude's network egress does not yet allow Strand's storage host. Strand's troubleshooting says a Claude organization administrator must add storage.googleapis.com under the allowed network egress setting, then you start a new chat. Everything except upload works without it. For multi-gigabyte whole-slide images, Strand recommends uploading in the browser at its web platform instead.

Are my slide images sent to the AI model?

No, not the pixels, according to Strand. Its security page says the in-chat Lattice viewer is read-only and that result pixels never enter the model's context. The model does see a bounded sample card and result summary, and core tools return expiring download links into the conversation. Strand asks users to keep those links out of prose, logs and screenshots.

Can I use Strand AI for clinical diagnosis?

No. Strand labels the connector research use only and states it is not for clinical diagnosis, treatment or patient-care decisions. Its docs say assistants are instructed to decline clinical requests. Slides must also be de-identified before upload, and Strand says its optional automated de-identification does not certify that data meets HIPAA or any other legal de-identification standard.

Sources

Use it in an agent

Put Strand AI to work.

Connect Strand AI once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.

Call (650) 285-1019, Agentman.Our AI receptionist answers.