OpenTacit is not only for coding harnesses. Each AI tool that can add a remote MCP connector can reach your organization's playbook. This includes plain chat with Claude and ChatGPT. You add one URL and sign in with your normal org account. Then the chat model can pull from OpenTacit in the conversation.

This is the lightest way to use OpenTacit: no binary, no API key, and no installation on your machine. It is also the correct path for surfaces where the full installation cannot run. Examples: Claude on the web and on phones, or a chat app that your organization has made standard.

What a connector gives you

A chat tool connector gives you the pull half of OpenTacit. This is the same half that you reach with @tacit and the /tacit: commands in a coding session:

Tool the chat model can call	What it does
tacit_search	Finds the techniques that are the best match for a query or for the current work
tacit_metrics	Shows how well the playbook performs. Its view parameter selects the funnel, the cohort breakdown, or the playbook map. On clients that support it, the result appears as an interactive panel
tacit_drafts / tacit_draft_action	Review and decide drafts, if your key has that permission

A connector does not give you the push half. The push half has two parts. One part is the unrequested ◆ suggestion that arrives in the middle of a task. The other part is the quiet, automatic capture of evidence from your session. These parts depend on the lifecycle hooks that a coding harness installs. A chat app does not expose these hooks. Thus, in a chat tool, OpenTacit responds only when the model asks. The only evidence that OpenTacit records is explicit: a tool call, or feedback that you give in words. If you want OpenTacit to observe your work and give suggestions without a request, wire a coding harness with tacit connect instead.

What your registry needs first

Chat connectors sign you in; they do not carry a member key. Thus your organization's registry must have sign-in (OIDC) turned on. A registry that uses key-only auth cannot accept a consumer chat connector. In that case, connect a coding harness with your member key. Or ask your administrator to configure sign-in.

You need one item: the registry's MCP address. This is the base URL with /mcp at the end.

https://longreach.tacit.zone/mcp

If your organization serves OpenTacit under a sub-path, keep the whole path: https://demo.example.com/apps/tacit/mcp.

Connect Claude (web, desktop, mobile)

1. Open Settings → Connectors. Select the option to add a custom connector.
2. Paste your registry's /mcp URL.
3. Sign in when Claude sends you to your organization's login. Use the same account that you use for the dashboard.

OpenTacit's tools are now available in each conversation. Ask in plain words, for example "search OpenTacit for deploying a lambda behind an api gateway" or "check this against our playbook". The model then calls the tool that matches. On Claude's web and desktop clients, the richer results (the drafts queue, the playbook map) appear as interactive panels. On other clients, they come back as text.

Connect ChatGPT

ChatGPT reaches OpenTacit through its Connectors surface. On some plans, it is necessary to turn on developer/MCP support first.

1. Add a connector and give it your registry's /mcp URL.
2. Complete the sign-in when ChatGPT asks.

Then ask ChatGPT to consult OpenTacit the same way. Results come back as text.

Menu names on both products change from release to release. If you cannot find "Connectors," look for "Custom connectors," "MCP servers," or a developer/beta settings area. The item that you add is always a remote MCP server URL with a sign-in.

Make it almost proactive

A chat tool cannot push. The nearest equivalent to "OpenTacit reminds me" is a reminder in a place where the model will read it on each turn. In a Claude Project or a Custom GPT, add an instruction such as:

Before you propose an approach, search our OpenTacit playbook for a validated move. Follow the move if the evidence supports it.

The model then uses OpenTacit on its own at the important moments. This is weaker than a hook, because the model decides when to look. But it makes a playbook search the default, not a manual step.

Make sure that it works

Ask the tool to run a search that you know has an answer: "search OpenTacit for <something your playbook covers>." A good result confirms the URL, the sign-in, and retrieval, end to end. If the model says that it has no OpenTacit tool, check the URL again. Make sure that the URL contains the sub-path, if there is one, and /mcp at the end. Also make sure that you completed the sign-in.

What stays private

The privacy line is the same line that the rest of OpenTacit obeys. OpenTacit uses the text of your query to find techniques, and it never stores the text. The registry reports everything by cohort, never by person. See How OpenTacit behaves in your harness for the full description. Your cohort comes from the account that you use to sign in. Thus reports go to the correct team, and you do not set anything.