How is this better than using a separate chat for criticism?

Sessions run in complete isolation. Your main conversation stays clean—no copy-pasting, no context pollution. You get the insight without the mess.

What makes this different from just prompting Claude harder?

We inject structured tools that create auditable reasoning trails. "Be critical" is a suggestion. Our tools leave evidence.

Why does ChatGPT always agree with me?

AI models are trained to be helpful and avoid conflict, leading to sycophantic behavior. They validate your ideas instead of challenging them. Our tools use structured tool calls that force systematic evaluation, preventing the AI from simply agreeing.

How does billing work?

$20/month or $180/year for full access to all reasoning tools. Cancel anytime from your account settings. No usage metering, no overage charges.

Who actually uses this?

Mostly developers and technical founders using AI coding agents (Claude Code, Cursor, Windsurf). When the agent is on iteration 30 of the wrong approach, an isolated reasoning session breaks the loop.

What do you do with my data?

Your reasoning sessions are private and encrypted. We do not train models on your data. We do not sell or share your content. Sessions are stored only for you to access and reference.

Won't this slow me down?

The reasoning step takes seconds. Rewriting a hallucinated backend takes weeks. The fastest way to ship is to not build the wrong thing.

Connect on claude.ai

One onboarding. Every Claude surface.

Add reasoning.services as Custom Connectors in claude.ai — they appear automatically in web, mobile, Cowork, Claude Desktop, and Claude Code (once you /login Claude Code to the same account). Four tools, four trips through the modal, paid once.

Open Claude.ai Connectors Cursor / Windsurf / Cline user?

Why bother with multiple trips? Once a connector is here, Anthropic mirrors it into every Claude client where you're signed in: web, mobile, Cowork, Claude Desktop, and Claude Code (after /login). Pay the friction once, get the tools everywhere.

Using a different client? Use the API key guide instead

You need a reasoning.services account

A paid account is required. No free tier — the connector authenticates via OAuth and checks your subscription on every call.

Setup guide

Complete these steps once per tool. Four tools = four passes through Steps 2–8.

Step 1 — Open the Connectors page

Claude.ai Connectors page showing Web and Not connected groups

Go to claude.ai/customize/connectors. You'll see your existing connectors grouped under “Web” and “Not connected.” If this is your first connector, both groups may be empty — that's fine.

Step 2 — Click the + (top right of the Connectors column)

Plus icon highlighted in the top-right corner of the Connectors column

You're looking for the small plus icon next to the search icon, not the one in the left sidebar. The latter creates a new chat.

Step 3 — Choose “Add custom connector”

Menu showing Browse connectors and Add custom connector options

Two options appear: Browse connectors (Anthropic's verified directory) and Add custom connector. reasoning.services lives in the custom path — pick the second.

Step 4 — The modal opens

Add custom connector modal with two empty input fields

A dialog titled “Add custom connector (Beta)” appears with two empty fields. The Beta tag is Claude.ai's, not ours — the underlying spec (MCP Streamable HTTP) is stable.

Step 5 — Paste a service name and URL

Add custom connector modal with Name and Remote MCP server URL fields filled

You're going to do this four times, once per tool. Each row below contains the exact Name and Remote MCP server URL to paste.

Name	Remote MCP server URL	Copy
Structured Reflection	`https://reasoning.services/tools/structured-reflection/mcp`
Sequential Thinking	`https://reasoning.services/tools/sequential-thinking/mcp`
Context Switcher	`https://reasoning.services/tools/context-switcher/mcp`
Decision Matrix	`https://reasoning.services/tools/decision-matrix/mcp`

Advanced settings is for self-hosted OAuth clients. Leave it closed — reasoning.services handles OAuth via Dynamic Client Registration automatically.

Step 6 — Click Add

Connector panel showing new connector in the Not connected group with Connect button visible

The modal closes. The new connector appears under “Not connected.” Right panel shows a Connect button. You're not done yet — the connector exists but Claude has no token to call it.

Step 7 — Click Connect → sign in to reasoning.services

A popup window opens to reasoning.services, not to Anthropic. You'll see two options:

Sign in with your corporate ID (GitHub) — for developer accounts
Continue with Google — for everyone else

This is the same sign-in you used when you created your reasoning.services account. The popup closes itself when authentication succeeds.

Phishing check

The URL bar in the popup should read *.reasoning.services or auth.reasoning.services. If it doesn't, close the popup and start over.

Step 8 — You're connected

Connector panel showing connector moved to the Web group with tool permission list visible

Back on the Connectors page, the new connector has moved into the Web group. The right panel shows its tools — typically split into “Read-only tools” and “Write/delete tools.” Default permission is Always allow.

Connection typically completes in under 30 seconds. If it takes longer, see Troubleshooting below.

Step 9 — (Optional) Customize tool permissions

Permission dropdown showing Always allow (selected), Needs approval, Blocked, and Custom

Click any Always allow dropdown to change the trust level for that tool category.

Always allow: Claude calls the tool without asking. No interruption to the reasoning loop.
Needs approval: Claude pauses and asks you before each tool invocation.
Blocked: Claude cannot call this tool at all. Use to disable temporarily without disconnecting.
Custom: Per-tool overrides within a category. Primarily useful for mixed-sensitivity connectors.

Step 9.5 — Our recommendation: leave Always allow everywhere

Always allow is correct for every reasoning.services tool. Here's why the alternatives are counterproductive:

Needs approval — Every approval prompt breaks the loop. The cognitive value of the tool comes from Claude reasoning with you in real time — interruption defeats the point.

Use this	When
Always allow (default)	Always, for reasoning.services tools
Needs approval	Only if auditing — switch back after
Blocked	Only to disable temporarily without disconnecting
Custom	Effectively reserved for mixed-sensitivity connectors — not useful here

Step 10 — Verify everywhere you use Claude

Your connectors are now available on every Claude.ai-authenticated surface: web, mobile, Cowork, Claude Desktop, and Claude Code (once you /login Claude Code to the same account).

Claude Code bridge

Confirm your Claude.ai account is active in Claude Code and verify the MCP servers appear:

bash

# Confirm Claude.ai is your active auth method
claude /status

# Open Claude Code and list MCP servers
claude
> /mcp

Expected output: entries prefixed with claude.ai for each reasoning.services connector, status connected.

If the connectors are missing, clear competing auth credentials and log in with your Claude.ai account:

bash

unset ANTHROPIC_API_KEY ANTHROPIC_AUTH_TOKEN
# remove any apiKeyHelper from ~/.claude/settings.json
claude /login        # pick your Claude.ai account

Now do this three more times.

Repeat Steps 2–8 with each URL from Step 5. Once all four connectors are connected:

Claude.ai web
Claude.ai mobile (iOS and Android)
Claude Cowork (shared workspaces)
Claude Desktop (synced from your account)
Claude Code — after claude /login with your Claude.ai account

§7.6 — Disconnecting or removing a connector

Two different actions live in the same Connectors panel. They look similar but have opposite consequences for your OAuth token and tool permissions.

Action	OAuth token	Tool permissions	When to use
Disconnect	Revoked — you re-authorize on next Connect	Preserved	Temporary — switching accounts, debugging auth
Remove	Revoked and deleted	Destroyed	Permanent — retiring the connector entirely
Re-connecting after Disconnect restores your saved permission settings.
Re-adding after Remove starts from scratch — Always allow everywhere is the default.

Warning

Anthropic's modal copy is identical for Disconnect and Remove. Both dialogs say “Are you sure?” with no additional context. Read the action label in the connector panel carefully before confirming.

To disconnect

1. Open the Connectors page. Your active connector is in the Web group on the left panel. Click it to select it.

Connector panel showing an active connector in the Web group with Connect and kebab menu visible

2. Click the “Disconnect” button in the right panel (or via the kebab menu). A confirmation dialog opens.

Disconnect confirmation dialog asking if you are sure you want to disconnect

3. Confirm. The connector moves back to “Not connected.” Your permission settings are preserved for when you reconnect.

Connector panel showing the connector moved to the Not connected group

To remove

1. Open the kebab menu (three dots) next to the connector name in the left panel. Select “Remove.”

Kebab menu open showing Disconnect and Remove options

2. A confirmation dialog opens. This dialog looks the same as the Disconnect dialog. Confirm only if you intend a permanent removal.

Remove confirmation dialog asking if you are sure you want to remove the connector

What happens to Claude Code

Claude Code mirrors the Claude.ai connector state via your account session.

Disconnect: Claude Code stops calling the tools immediately. Run /mcp to confirm the server shows as disconnected. Reconnect on claude.ai and it re-activates.
Remove: The server entry disappears from /mcp entirely. You must add the connector again from scratch.

Troubleshooting

Common issues and how to resolve them.

Connector stuck in "Not connected" after clicking Connect

The OAuth popup may have been blocked. Allow popups for claude.ai in your browser settings, then click Connect again.

Popup opens but closes immediately without success

Your browser may have third-party cookie restrictions. Try a different browser or temporarily allow cookies from reasoning.services.

Tools appear in the panel but Claude doesn't use them

Tool permissions may be set to Blocked. Open the connector detail, set all tool categories to Always allow, and try again.

"Authorization failed" or "Invalid token" error

Your session may have expired. Click Disconnect, then Connect again to re-authorize. If the error persists, remove and re-add the connector.

Connector doesn't appear in Claude Code /mcp

Claude Code must be logged in to the same Claude.ai account. Run claude /login and select your Claude.ai account. Then run /mcp to verify.

Added the connector but only see some tools

Each reasoning.services service is a separate connector. You need four connectors total — one per tool. Repeat Steps 2–8 for each URL in Step 5.

Connection succeeds but tools disappear after a day

Your OAuth token expired and was not refreshed. This is rare — if it recurs, disconnect and reconnect to re-issue a token.

Wrong account connected

Click Disconnect. When you click Connect again, the OAuth popup will let you sign in with a different account.

Ready to add the tools?

Four connectors. Four trips through the modal. Reasoning on every Claude surface.

Open Claude.ai Connectors Sign up first

Navigation