LatchBio
by LatchBio
Browse bioinformatics data, launch Nextflow and Python workflows, and monitor executions from an AI agent. 9 tools, matching Anthropic's listing exactly. OAuth sign-in, identity-only scopes, and one tool that spends money.
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.
Connect LatchBio via MCP
https://mcp.latch.bio/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
LatchBio Tools & Capabilities (9)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- One tool spends money, and no scope prevents it.
launch_workflowbills your workspace at published Latch rates. The only controls are your MCP client's approval prompt and your Latch workspace role — not the OAuth grant, which draws no line between reading and launching. - Nothing stops, cancels or deletes. The nine tools cannot cancel a running execution, delete a file, or remove a workflow. A run started through the connector must be stopped from the Latch Console or SDK. That asymmetry cuts both ways: it limits blast radius, and it means the agent cannot clean up after itself.
- No plan gate, but a billing prerequisite. LatchBio charges no MCP-specific fee, and its pricing model has no subscription or per-seat charge. But launching anything requires credits, so a workspace without a payment method can browse and monitor and not much else.
- Public workflows are not vetted.
list_workflowssurfaces community-built workflows alongside Latch Verified ones, and LatchBio states community workflows are not actively vetted. An agent choosing from that list is choosing which container to run on your data. - Presigned URLs land in your transcript.
get_fileandget_task_logscan return download URLs that work for anyone holding them. LatchBio does not publish their expiry, and we did not test it. Treat them as live credentials of unknown lifetime. - We could not read tool schemas or safety annotations. The endpoint returned 401 to an anonymous request, so no parameter-level detail — including what
launch_workflowaccepts, or whether it can request a specific instance size — is published here. - No published MCP rate limits. LatchBio documents no rate limits for the MCP server, and we did not probe for them. Absence of documentation is not absence of limits.
- Your data leaves Latch's guarantees when a tool returns. LatchBio commits not to train models on your data and holds SOC 2 Type II. Neither commitment follows a tool result into your AI provider's context. For human genomic or clinical data, that is the decision to make before connecting.
trust.latch.biowas not readable. LatchBio's security page directs readers to its Trust Center for its 50-plus safeguards; that host returned 403 to our request on 2026-08-22, so the compliance claims cited here come from LatchBio's own FAQ and marketing pages rather than from audit artefacts we read.
Frequently asked questions
Yes, at the same rates as launching it yourself. LatchBio's docs state you are charged normally for any action performed through the MCP, the same as from the Console or SDK. Its pricing page sets 1 credit at 1 US dollar and gives a worked bulk RNA-seq run at 14.101 credits for 18 minutes.
LatchBio states that by default Claude will ask permission before launching workflow executions. That default lives in your MCP client, not in the server, so it is a client setting you can change rather than a server-side block. Treat approval as the control and keep it on.
Only identity ones. The server's RFC 9728 descriptor declared scopes_supported of openid and offline_access on 2026-08-22, which identify you and keep the session alive. Neither narrows anything. No scope separates listing a file from launching a workflow that spends credits, so consent is all-or-nothing.
Yes, if your workspace holds it. Latch Data stores FASTQ, BAM and single-cell files, and LatchBio's own documentation works through examples using human patient cohorts. Whatever a tool returns enters your AI client's context, so the connector's read reach is exactly your workspace's data, not a filtered subset.
Effectively yes. The get_file tool returns access information as either a Latch Console link or a presigned download URL, and get_task_logs can also return one. A presigned URL is a bearer credential — anyone holding the string can fetch the file without signing in to Latch.
No. None of the nine tools carries a delete, remove, cancel, abort or stop verb, and LatchBio's own tool table describes none. The surface writes in exactly one direction: launch_workflow starts new work. Stopping or deleting anything still requires the Latch Console or SDK.
Yes. LatchBio's FAQ states the company is SOC 2 Type II compliant following a three-month independent audit that included penetration tests. Its security page adds annual penetration tests, AES-256 encryption at rest and in transit via AWS KMS, and role-based workspace access assigned by administrators.
LatchBio documents four by name: Claude web and Desktop as a third-party connector, Claude Code, Codex and Cursor, each over OAuth at the same URL. Anthropic's directory lists compatibility with Claude, the Claude API and Claude Code. Any other MCP client can use the same endpoint.
Sources
- LatchBio MCP documentation (retrieved 2026-08-22). Fetched as Markdown via the
.mdsuffix.wiki.latch.bio/robots.txtcarriesContent-Signal: ai-train=yes, search=yes, ai-input=yes— LatchBio expressly permits AI synthesis of this content. · retrieved 2026-08-22 - LatchBio documentation index,
llms.txtandllms-full.txt— · (retrieved 2026-08-22) · retrieved 2026-08-22 - LatchBio pricing, credit rates and worked bulk RNA-seq example (retrieved 2026-08-22) · retrieved 2026-08-22
- LatchBio Security & Compliance (retrieved 2026-08-22) · retrieved 2026-08-22
- LatchBio FAQ, data policy and SOC 2 Type II (retrieved 2026-08-22) · retrieved 2026-08-22
- LatchBio privacy policy, effective 13 May 2026 (retrieved 2026-08-22) · retrieved 2026-08-22
- Latch Workflows Console documentation, Verified versus community workflows (retrieved 2026-08-22) · retrieved 2026-08-22
- Latch workspace roles (retrieved 2026-08-22) · retrieved 2026-08-22
- Latch Data file types and visualizations (retrieved 2026-08-22) · retrieved 2026-08-22
- Live OAuth posture check: anonymous
initializereturned 401 with awww-authenticatechallenge naming the resource metadata document; RFC 9728 descriptor athttps://mcp.latch.bio/.well-known/oauth-protected-resource/mcp; authorization server metadata athttps://auth.latch.bio/.well-known/oauth-authorization-server(2026-08-22) · retrieved 2026-08-22 - Anthropic Connectors Directory entry — , read from our directory snapshot dated 2026-08-16 · retrieved 2026-08-16
- LatchBio support — <mailto:support@latch.bio> · Privacy
Use in Agentman
Connect once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioServer Info
- Category
- Healthcare
- Developer
- LatchBio
- Tools
- 9
- Domain
- mcp.latch.bio
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.