Use Fostera AI Memory in Claude, ChatGPT, Gemini and Grok

Fostera's MCP connector lets Claude, ChatGPT, Gemini and Grok request context from a Soul you authorize. A Soul is your persistent AI identity and memory in Fostera. MCP—the Model Context Protocol—is the connection that lets another AI app use that context while you work there.

For example, you can ask your connected app to recall your preferred writing style before drafting a client update. The app requests relevant information from Fostera and uses the result in its answer. You can inspect the tool call and manage the connection in Fostera.

The server address is https://mcp.fostera.ai/mcp. Use the connection guide for a copy button, direct setup links, current access details and troubleshooting.

What a connection does

A read-only Soul connection can retrieve the selected Soul's identity, memories, memory summary, projects, creation text, familiarity and ability descriptions. It does not give the app permission to change or delete your Soul, execute abilities, or read every chat in your account. The consent screen states exactly what a connection can access.

Start with a concrete task:

Use Fostera to recall my preferred writing style before drafting this client update. Tell me which context you used.

A useful answer should show an actual Fostera tool call. An assistant saying “I remember” without using the connector is not proof that it retrieved your Soul's memory. Check the app's activity or tool details if you are unsure.

Importing and connecting solve different problems

Connecting Fostera does not grant access to all the chats in your Claude, ChatGPT, Gemini or Grok account. The host app must supply the content you want to import, or you must provide a supported export. A conversation summary is not a verbatim transcript. No connector can reconstruct messages or attachments missing from the source.

The history import walkthrough explains provider exports and how to review derived memories. Check the current import preview before committing: the selected import mode determines what Fostera stores.

Connect Claude

1. Open Claude with Fostera's name and URL filled in.
2. Review the custom connector and add it. If the shortcut does not open the form, go to Customize → Connectors → Add custom connector, name it Fostera, and paste https://mcp.fostera.ai/mcp.
3. Select Connect next to Fostera.
4. Sign in to Fostera, choose your Soul, review access and select Allow.
5. Open a conversation, enable Fostera in Claude's tools, and try the prompt below.

The shortcut prepares the setup form; it does not authorize access for you. Anthropic documents prefilled custom connector links and custom connector setup. Organization administrators may need to make a connector available first.

Connect ChatGPT

1. Open ChatGPT Plugins.
2. Select +, then Add custom MCP server.
3. Name it Fostera, enter https://mcp.fostera.ai/mcp, and choose OAuth authentication.
4. Review ChatGPT's custom-server notice and create the plugin.
5. Sign in to Fostera, choose your Soul, review access and select Allow.
6. Start a chat with Fostera enabled and try the prompt below.

This link opens Plugins; it does not prefill or install the connector. Account and workspace policies determine whether the custom-server option appears. Follow OpenAI's current MCP connection instructions if your menu differs.

Connect Gemini

1. Open Gemini on a computer.
2. Open Settings → Connected Apps; this may appear under Personal Intelligence.
3. Under Custom apps, select Add a custom app.
4. Enter https://mcp.fostera.ai/mcp, leave Advanced features closed, and continue.
5. Sign in to Fostera, choose your Soul, review access and select Allow.
6. In a conversation, type @, choose Fostera, and try the prompt below.

The link opens Gemini, so you still navigate to the custom-app form. Google's custom app requirements apply, including account, region, age, language and activity settings. If custom apps are unavailable on your account, a direct link cannot enable them.

Connect Grok

1. Open Grok Connectors.
2. Select New Connector → Custom.
3. Name it Fostera and enter https://mcp.fostera.ai/mcp.
4. Continue to Fostera, sign in, choose your Soul, review access and select Allow.
5. Return to Grok and ask it to use Fostera explicitly.

The link opens connector settings; it does not authorize a connection. See Grok's connector documentation for current app and organization requirements.

Verify your first connection

Use this first prompt in any connected app:

Use Fostera to identify my connected Soul and summarize its available memories. Tell me which Fostera tools you used.

Check that the tool result names the Soul you selected. Then ask a narrow question about a memory you can verify in Fostera. If the assistant cannot find the connector, enable it in the conversation's tools menu and try a new chat. If sign-in fails, confirm that the URL ends in /mcp and allow the browser's sign-in window.

Beyond reading: separately authorized Soul tools

Fostera also supports an expanded MCP toolset behind its advanced-platform rollout. These tools appear only where that rollout is enabled and the connection has the required permissions. The standard connection remains read-only; existing grants do not expand automatically.

With explicit authorization, an AI app can create a Soul, edit the selected Soul, browse its chats, search supported app scopes, or export chat text and project manifests. Account-wide search and reading your Resonance balance or usage require their own account permissions. Soul deletion opens a separate review in Fostera. Ability execution likewise requires review of the exact operation and arguments in Fostera, and follows the usual permission, approval and billing rules. Discovering an ability is different from authorizing it to run.

For imports, the assistant supplies text you have chosen or a supported export. Fostera previews the exact content before storing it as ordinary Soul chats and projects. Imported messages carry source labels and are treated as quoted history, not active system instructions. Storing transcripts does not automatically extract memories. The existing memory import flow remains a separate choice.

Where enabled, open the transcript importer to select an existing Soul, read a JSON export locally, choose conversations, review the preview, and confirm storage. A source file can be up to 5 MiB; each stored batch is limited to 20 conversations, 300 messages and 512 KB. For larger archives, continue with the next reviewed batch: the browser carries forward the imported project mapping within that archive. Each preview says whether it creates or continues a project. Large conversations are offered as consecutive numbered message ranges; each range becomes a separate chat, and continuing until no ranges remain retains every supplied message. Individual messages must fit the 64,000-character message limit; split an oversized message explicitly in portable JSON before importing. Keep your original archive. Provider adapters may omit unsupported roles, missing responses, attachments or native project membership. Portable Fostera JSON can preserve supplied project titles and conversation membership, but does not recreate a provider’s private settings or files.

MCP exports are paginated: follow each cursor to retrieve the complete chat text. A project export is a manifest of its chats; export those chats individually for their text. This is not a promise that another provider accepts Fostera’s format through its native import UI. Your connected assistant can use the exported context where its host permits.

For MCP clients, pass continueProjectId on a portable project only after reviewing the project ID returned in an earlier batch’s projectMappings. Fostera checks ownership, Soul, source and original project identity; it never merges unrelated projects just because their source IDs match. Undo removes only projects created by that batch, and refuses to remove a project containing another batch’s chats.

An import has a stable request key so retrying it does not create duplicate chats. After completion, you can undo the imported batch while it remains unchanged. Later messages or other changes prevent that shortcut; use the ordinary chat and project controls to manage later work.

To add permissions, use a host that can request the corresponding scopes, reconnect, and review the new request in Fostera. Reconnecting a host that requests only read access does not unlock write tools. Check the live connection guide for the capabilities currently available.

Keep access under your control

Open Settings → Models & keys → Connected AI apps in Fostera to review or disconnect an app. Open connected-app settings. Disconnecting stops future access within 30 seconds; it cannot retract information the other app already received.

Connections do not silently gain new powers when Fostera adds tools. Existing read-only connections remain read-only. If you need a capability outside the permissions shown, review the available connection options and explicitly authorize it. An assistant cannot grant itself extra access by retrying a tool.

Use separate Souls and project boundaries for context that should stay separate. Before connecting client material, consider which AI app will receive it and review that app's data policies alongside Fostera's privacy policy.

Questions about MCP and AI memory

Does connecting move my entire account?

No. It gives the app access to the Fostera tools and data you authorize. Importing history is a separate, reviewed operation. Source-provider exports may omit attachments, branches or messages.

Can I use the same Soul from different AI apps?

Yes. Connect each app separately and select that Soul. Each connection has its own permissions and can be disconnected independently.

Is an ability description the same as permission to run it?

No. Reading a Soul's abilities tells the app what the Soul can do. Execution needs an available execution tool, appropriate permission and any required confirmation. A read-only connection does not provide that permission.

Does MCP guarantee the assistant uses my memories?

No. The host app decides when to call a tool. Ask it to use Fostera explicitly and inspect the tool result when context matters. Retrieved memories can also be incomplete or outdated; correct them in Fostera.

Can I disconnect without deleting my Soul?

Yes. Disconnect the app in Fostera's connected-app settings. Your Soul remains in Fostera. Manage any retained copy in the other app separately.

Start with one connection, one Soul and a task whose context you can verify. If your useful context is still in another platform, import selected history first.