Connect Niyra to Claude, ChatGPT, or any MCP client

When you connect Niyra to another AI assistant, your Niyra becomes a tool inside it. Claude or ChatGPT can then reach into your world — your email, calendar, documents, long-term memory, and commitments — through Niyra, using your integrations and gated by permissions you approve. You stay in the tool you like; that tool stops being a blank-slate chatbot and starts knowing you.

Niyra speaks the Model Context Protocol (MCP) over HTTPS, so it works with any MCP-compatible client.

Everything you need

| | |
|---|---|
| MCP endpoint | https://api.niyra.ai/mcp |
| Protocol | JSON-RPC 2.0 · MCP 2025-06-18 |
| Auth | OAuth 2.1 (sign in + approve) — or a Personal Access Token |
| OAuth discovery | https://api.niyra.ai/.well-known/oauth-authorization-server |
| Resource metadata | https://api.niyra.ai/.well-known/oauth-protected-resource |
| JWKS | https://api.niyra.ai/.well-known/jwks.json |
| GPT-Action spec | https://api.niyra.ai/v1/public/openapi.json |

For most tools, the endpoint URL is the only thing you paste. Everything else — sign-in, permissions — is discovered automatically.

How signing in works

You never hand a tool a raw password. When a client connects for the first time:

1. It opens Niyra's authorization page at api.niyra.ai/oauth/authorize.
2. If you're not signed in, Niyra sends you to sign in (the same login you use for niyra.ai) and brings you right back.
3. You see a consent screen showing which app is asking and exactly what it wants to do (ask questions, read memories, take actions…). You Approve or Deny.
4. On approve, the tool is connected. It never sees your password — only a scoped token Niyra issues.

You can review or revoke any connected app anytime in Settings → API → Authorized integrations.

Connect your tool

Claude Desktop / claude.ai

1. Open Settings → Connectors → Add custom connector.
2. Paste https://api.niyra.ai/mcp and name it Niyra.
3. Sign in + approve when prompted.

> Shortcut: Niyra's own Settings → API page has a one-click "Add to Claude" button that pre-fills this.

Claude Code

Run in your terminal:

Cursor

Settings → MCP → Add server, URL https://api.niyra.ai/mcp. (The Settings → API page also has a one-click Cursor deep link.)

ChatGPT

Two ways, pick one:

- MCP connector (Developer mode / Deep Research): add https://api.niyra.ai/mcp as a connector and sign in.
- Custom GPT Action: in the GPT builder, import the schema from https://api.niyra.ai/v1/public/openapi.json and set auth to OAuth (or paste a Personal Access Token as a Bearer key).

Cline / Windsurf / Continue (and any other MCP client)

Paste this into the tool's MCP config:

What Niyra can do once connected

The other assistant will see these tools:

| Tool | Permission | What it does |
|---|---|---|
| niyraask | niyra:ask | Ask a question that needs your memory, email, calendar, or docs — read-only |
| niyramemories | niyra:memory:read | Search what Niyra remembers about you |
| niyraremember | niyra:memory:write | Save a new fact or preference |
| niyraexecute | niyra:execute | Ask Niyra to do something (send an email, schedule a meeting) — destructive steps still confirm in your own channel |
| niyrafollowup | niyra:ask | Continue an ongoing thread or task |
| niyragettask | niyra:ask | Check on a long-running task Niyra spun off |
| niyracommitments | niyra:memory:read | See what Niyra has flagged that you owe people |

Try these once you're connected

- "Ask my Niyra what meetings I have this week."
- "Ask Niyra to summarize my last 10 emails from the legal team."
- "What does Niyra remember about my priorities?"
- "Have Niyra remember I prefer morning meetings."
- "Ask Niyra to draft a reply to the last email from Sarah." (needs the execute permission)
- "What has Niyra flagged that I owe people?"

Tip: when in doubt, just say "ask Niyra …" — that routes the request to niyraask.

Permissions (scopes)

Niyra issues fine-grained, per-connection permissions. You approve them once at consent time; changing them means re-authorizing.

| Scope | Grants |
|---|---|
| niyra:ask | Read-only questions and lookups |
| niyra:memory:read | Read saved facts, preferences, history |
| niyra:memory:write | Save new facts and preferences |
| niyra:resources:read | Read conversation transcripts and task results |
| niyra:tools:safe | Use read-only tools (web search, calendar lookups) |
| niyra:execute | Run action loops (send, schedule, mutate) |
| niyra:tools:destructive | Required for destructive actions — still confirmed by you |

Full detail: Scopes reference · OAuth flow.

Prefer a manual token?

For scripts or tools that want a plain bearer instead of the OAuth flow, mint a Personal Access Token in Settings → API. Then send:

Troubleshooting

- "Not authorized" / repeated sign-in loop — make sure you're logging into the same Niyra account you use on niyra.ai, and that pop-ups aren't blocked during the OAuth redirect.
- Tool connects but answers are empty — you're on an old build; the ask lane requires the current release. Reconnect, or reach out at founders@niyra.ai.
- Client won't auto-discover the login — a few strict clients need the resource-metadata URL directly: https://api.niyra.ai/.well-known/oauth-protected-resource.
- Actions do nothing — the connection likely only has read scopes. Re-authorize and approve the execute permission.

Still stuck? Email founders@niyra.ai** and we'll get you connected.