Lattice

Name: Lattice MCP Server
Availability: InStock
Author: Lattice

Productivity20 tools

Draft, validate and submit performance reviews from your AI assistant. Lattice built a genuine draft-validate-submit pipeline, and submission locks the review: reversing it needs an admin, not another tool call.

Verified connector

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

Connection checked by Agentman on 19 August 2026.

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

Connect Lattice via MCP

https://api.latticehq.com/mcp

Works in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.

Lattice Tools & Capabilities (20)

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

Limits

submit-review finalises. Lattice states the review transitions drafted → submitted and is then no longer editable as a draft.
No undo in the connector. No listed tool reopens, unsubmits, edits or deletes a submitted review. Reopening is an in-product request an admin approves, or an admin action.
Reopening is not quiet. Lattice states the reviewer is notified when an admin reopens a review and that this notification is not optional; the previously submitted version is hidden from managers and peers until resubmission.
After packet sharing, a reopen cannot be requested. Lattice states you must contact an admin instead.
Admins may already have read it. Lattice states admins can view all reviews as soon as they are submitted, and can view drafts through impersonation.
Partial reads look like complete ones. Lattice states read requests return only what you may see, sometimes partially — the response does not announce what was withheld.
Two directory tool names are undocumented. get-review-packet-for-reviewee and get-review-context-for-reviewee appear in Anthropic's snapshot and in no Lattice article we found, so what a review packet contains is not documented for MCP.
Two more names differ by a trailing letter between the snapshot and Lattice's tools article; on a kebab-case server that is a hard mismatch.
The tool list is explicitly unstable. Lattice states it changes over time and names the client's own tool listing as authoritative.
No administrator capabilities. Lattice states admin-only data and actions are unreachable through MCP, for every user including super admins.
No dynamic client registration. Stated by Lattice and absent from both metadata documents.
Two advertised write scopes have no matching tool. mcp:feedback.write and mcp:meetings.write exist in the descriptor with no write tool in either source.
Two admins must act before anyone connects, and the Claude side requires a Team or Enterprise plan with Owner or Primary Owner rights.
Data residency documentation conflicts. The security article says US-only; the more recently updated FAQ says North America and EMEA.
PHI is prohibited. Lattice states its platform is not designed for storing Protected Health Information and that its terms prohibit entering it — which applies to anything an agent writes too.
Tool data depends on which Lattice products you bought. Goals, Grow and Performance are separately priced packages.
No prompts. Anthropic's snapshot lists none; Lattice's prompt library is help-centre prose you paste yourself.
We did not read the live schemas. The endpoint returned 401 to an anonymous handshake and we did not authenticate. Tool names come from Anthropic's snapshot and Lattice's documentation, not from a wire response — so no safety annotations were read, and none of these tools can be described here as annotated read-only.

Frequently asked questions

Does the Lattice MCP server actually submit performance reviews?

Yes. Lattice documents submit-review as transitioning a drafted review to submitted, after which it is no longer editable as a draft. It is one of three write tools and the only one that finalises anything. Two separate tools handle the earlier stages — save-draft-review writes answers without submitting, and validate-draft-review only reports what is missing.

Can you unsubmit a review submitted through Lattice MCP?

Not through the connector. None of the twenty tools reopens, edits or deletes a submitted review, so an agent that submits cannot undo it. Lattice supports reopening in the product: a reviewer requests a reopen and an admin approves or denies it, or an admin reopens it directly. Lattice states the reviewer is always notified when that happens.

Whose performance reviews can Lattice MCP read and write?

Only what the signed-in employee could already see in Lattice. Lattice states the MCP server has no separate permission model and applies the platform's own roles, and that administrator capabilities are excluded entirely. It gives a worked example: a super admin calling find-reviews gets back their own reviews and their reports' reviews, not everyone's.

Does Lattice MCP expose the company org chart?

Yes, within your existing Lattice visibility. Three read tools cover reporting structure: find-reports returns an employee's subtree up to five levels deep, find-ancestors returns the management chain up to the org root, and list-departments lists active departments. Each returns names, titles, emails and entity IDs, so reporting lines become readable by an agent in bulk.

Can you connect Lattice MCP as read-only?

Yes, in two independent ways. Lattice's OAuth resource descriptor advertises eleven named scopes split by domain and direction, including separate mcp:reviews.read and mcp:reviews.write, so a token can carry reads without writes. Separately, Lattice documents that a Claude organisation owner can restrict the connector's tool permissions to read only.

Who has to enable Lattice MCP before employees can use it?

Two administrators, in sequence. A Lattice admin must set Claude to Allowed under Admin, Platform, MCP Integrations. Then a Claude organisation Owner or Primary Owner on a Team or Enterprise plan adds the connector for the team. Only after both steps can an individual employee connect, entering their Lattice subdomain and signing in.

How many tools does the Lattice MCP server expose?

Anthropic's directory snapshot lists twenty tool names; Lattice's own tools article documents seventeen. Two snapshot names appear in no Lattice article, and two more differ from the documented names by a single trailing letter. Lattice states the list changes over time and names the client's own tool listing as the authoritative source.

Sources

Set up Lattice MCP with Claude — https://help.lattice.com/hc/en-us/articles/40952435311127-Set-up-Lattice-MCP-with-Claude (fetched 2026-08-19 via the help centre's JSON API at /api/v2/help_center/en-us/articles/40952435311127.json; article last updated 2026-08-19. This is the documentation URL in Anthropic's directory). Appending .md to a Zendesk help-centre URL is ignored and serves HTML, so the JSON API was used throughout · retrieved 2026-08-19
Lattice MCP Tools — https://help.lattice.com/hc/en-us/articles/41580674227351-Lattice-MCP-Tools (fetched 2026-08-19; last updated 2026-08-19; 17 tools with type, description and example prompts, and the statement that the client's tool listing is authoritative) · retrieved 2026-08-19
Lattice MCP Security and Permissions — https://help.lattice.com/hc/en-us/articles/41580385880343-Lattice-MCP-Security-and-Permissions (fetched 2026-08-19; last updated 2026-07-08; no MCP-specific permissions, no admin capabilities, read vs write denial behaviour, PHI prohibition, US-only residency statement) · retrieved 2026-08-19
Lattice MCP Server FAQs — https://help.lattice.com/hc/en-us/articles/41618362331671-Lattice-MCP-Server-FAQs (fetched 2026-08-19; last updated 2026-08-19; no DCR, PKCE vs client secret, private 1:1 notes, North America and EMEA residency) · retrieved 2026-08-19
Lattice MCP Integrations Overview — https://help.lattice.com/hc/en-us/articles/40935994775831-Lattice-MCP-Integrations-Overview (fetched 2026-08-19) · retrieved 2026-08-19
Set up a Custom Lattice MCP Server — https://help.lattice.com/hc/en-us/articles/41581931712151-Set-up-a-Custom-Lattice-MCP-Server (fetched 2026-08-19; admin-created client ID and optional secret, redirect URLs) · retrieved 2026-08-19
Lattice MCP Prompt Library — https://help.lattice.com/hc/en-us/articles/41067604753431-Lattice-MCP-Prompt-Library (fetched 2026-08-19; includes both a wait-for-approval drafting prompt and a draft-and-submit prompt) · retrieved 2026-08-19
How Can I Edit My Review? — https://help.lattice.com/hc/en-us/articles/4404785571479-How-Can-I-Edit-My-Review (fetched 2026-08-19; editable until submitted, then reopen request required) · retrieved 2026-08-19
Request to Reopen Your Review — https://help.lattice.com/hc/en-us/articles/13772091375895-Request-to-Reopen-Your-Review (fetched 2026-08-19; reopened reviews hidden from managers and peers; no request possible after packet sharing) · retrieved 2026-08-19
Reopen a Review — https://help.lattice.com/hc/en-us/articles/1500001889261-Reopen-a-Review (fetched 2026-08-19; admin action, reviewer notification not optional) · retrieved 2026-08-19
When does my review become visible to others? — https://help.lattice.com/hc/en-us/articles/4404271997847-When-does-my-review-become-visible-to-others (fetched 2026-08-19; never auto-shared, admins see on submission, cycle deliverables govern the rest) · retrieved 2026-08-19
Lattice MCP product page — https://lattice.com/platform/mcp (fetched 2026-08-19) · retrieved 2026-08-19
Lattice pricing — https://lattice.com/pricing (fetched 2026-08-19; Foundations $13, Performance $10, Goals & OKRs $8, Engagement $4, Compensation +$6, Grow +$4 per seat per month; $4,000 minimum annual agreement) · retrieved 2026-08-19
help.lattice.com/robots.txt returns 200 from Zendesk with no Content-Signal header; help.lattice.com/llms.txt returns 404. lattice.com/robots.txt returns 200 with four disallow rules and a sitemap, again no Content-Signal; lattice.com/llms.txt returns 200 and the sitemap at lattice.com/sitemap.xml located the MCP product page (all fetched 2026-08-19) · retrieved 2026-08-19
Live RFC 9728 protected-resource descriptor — https://api.latticehq.com/.well-known/oauth-protected-resource/mcp (fetched 2026-08-19; 11 named scopes, one authorization server, JWKS URI, header bearer methods. Root and path-insert forms both return JSON 404s) · retrieved 2026-08-19
Live RFC 8414 authorization-server metadata — https://api.latticehq.com/.well-known/oauth-authorization-server/mcp (fetched 2026-08-19; same 11 scopes, S256 PKCE, no registration endpoint) and https://api-auth.latticehq.com/.well-known/oauth-authorization-server (fetched 2026-08-19; different issuer https://partner-auth.latticehq.com, only openid/offline/offline_access scopes) · retrieved 2026-08-19
Live anonymous probe — 2026-08-19: an anonymous initialize POST to https://api.latticehq.com/mcp returns HTTP 401 with www-authenticate: Bearer resource_metadata="https://api.latticehq.com/.well-known/oauth-protected-resource/mcp". No tool was ever called and no OAuth flow was ever begun · retrieved 2026-08-19
Anthropic connector directory — https://claude.ai/directory/9830ad9d-3a53-469b-a45c-aef16b059aa5 (snapshot 2026-08-16; permissions "Read and write", 20 tool names, no prompts listed, partner tier) · retrieved 2026-08-16