MCP server

Give ChatGPT, Codex, Claude Code, Cursor, VS Code, and other MCP clients governed access to Product Context and feedback.

View .md Open in Claude Open in ChatGPT

iterateTo serves standards-based streamable HTTP at https://iterateto.com/api/mcp. Agents can retrieve a bounded Product Context bundle, search it, inspect individual records, and—when granted the right scope—create, update, or archive records. Feedback tools remain available on the same connection.

Connect

For a local agent or IDE, create an itr_ personal access token under Dashboard → Settings → API keys, expose it as ITERATETO_API_KEY to the client process, then let the CLI write a project-scoped config. For hosted clients such as ChatGPT, add https://iterateto.com/api/mcp as a remote MCP server and sign in with OAuth; no API key is pasted into ChatGPT.

ChatGPT

In ChatGPT Developer mode, open Plugins, select the plus button, and use these exact connection values. ChatGPT discovers iterateTo's OAuth login from the endpoint; you do not create or paste a personal key.

ChatGPT field	Choose
Transport	Streaming / Streamable HTTP
Server URL	https://iterateto.com/api/mcp
Authentication	OAuth / automatic sign-in
Bearer token	Leave blank

Local coding tools

Add to Cursor Add to VS Code

Claude Code

bash

npx iterateto mcp init --client claude --host https://iterateto.com

Codex

bash

npx iterateto mcp init --client codex --host https://iterateto.com

Generated configuration references ITERATETO_API_KEY. Set that variable in the process which launches your client; never paste an itr_ secret into a committed config or install link.

Connect with the CLI

iterateto mcp init --client claude|codex|cursor|vscode safely merges the iterateTo server into the client config. Use --env-var NAME when your team uses a different secret variable.

bash

export ITERATETO_API_KEY=itr_your_secret_key
npx iterateto mcp init --client codex --host https://iterateto.com

To teach an agent the secure connect-and-verify workflow, install the optional cross-client skill: npx skills add https://iterateto.com/mcp-skill.md.

Tools

Tool	Description
`product_context_list_options`	Resolve the live member, team, product-area, and current-actor IDs required by write payloads.
`product_context_get`	Retrieve a bounded project context bundle, optionally filtered by record type.
`product_context_search`	Search context semantically when embeddings are configured, with deterministic lexical fallback.
`product_context_item_get`	Fetch one strategic artifact, outcome, opportunity, or solution-discovery record.
`product_context_create`	Create a validated Product Context record with an idempotency key.
`product_context_update`	Update a record with an idempotency key and optimistic version check.
`product_context_archive`	Archive one record; requires the administrative scope.
`feedback_list_projects`	List the projects in your organization.
`feedback_list`	List and filter feedback: project, status, kind, text search, tag; cursor-paginated.
`feedback_get`	Fetch one item in full — targets with selector candidates, notes, screenshot metadata.
`feedback_update`	Update status, priority, or tags (idempotent write).
`feedback_add_note`	Add an internal note with an idempotency key (≤ 2000 chars; REST allows up to 4000).
`feedback_stats`	Counts by status and kind, plus a per-day series.

Product Context resource

Clients that support MCP resources can read iterateto://product/{projectId}/context. The resource and tools use the same tenant checks and return only records belonging to the authenticated organization and requested project.

Authentication & scopes

The endpoint accepts OAuth 2.1 access tokens and itr_ personal access tokens. OAuth advertises its authorization-server and protected-resource metadata automatically, including PKCE and dynamic client registration for compatible remote clients. Hosted OAuth connections default to read-and-manage access; product:admin is never granted by default. Personal tokens can be read-only, read-and-manage, or full access. Available Product Context scopes are product:read, product:write, and product:admin; feedback has matching read/write scopes. Tools outside the token's scopes are not advertised.

Vector retrieval is optional. Set AI_GATEWAY_API_KEY and PRODUCT_CONTEXT_EMBEDDING_MODEL to use AI SDK embeddings with an Atlas Vector Search index. If embeddings or the index are unavailable, search explicitly reports that it used lexical fallback rather than pretending the result was semantic.

Ready-made agent workflows live in Recipes.