MCP Server - Nous

The Nous MCP server gives your agent the context graph as a focused set of tools, engineered and sourced context ready to act on. Have your agent call get_workspace_status first. It can set up the workspace itself.

Fastest setup

One command installs the CLI, signs you in, and registers the MCP with your agent.

curl -fsSL https://api.opennous.cloud/install | sh

Then tell your agent “set up my Nous workspace”. It finds your ICP, or drafts one from your website, and syncs it. That is onboarding. Prefer to run it yourself.

npx @opennous/cli init

init signs you in through the browser, saves a workspace key to ~/.nous/config.json, and registers the server with your agent.

Connect any client

Run npx @opennous/cli login once. It saves your key to ~/.nous/config.json, so no key goes in your config. Then add Nous to your client.

{
  "mcpServers": {
    "nous": { "command": "npx", "args": ["-y", "@opennous/mcp"] }
  }
}

Client	Where to add it
Claude Code	`claude mcp add nous -- npx -y @opennous/mcp`
Cursor	`~/.cursor/mcp.json`
Claude Desktop	`~/Library/Application Support/Claude/claude_desktop_config.json` (macOS). `%APPDATA%\Claude\claude_desktop_config.json` (Windows)
Codex	`~/.codex/config.toml`. TOML, shown below
Hermes	MCP settings. Set the name to `nous`, then paste the inner object `{ "command": "npx", "args": ["-y", "@opennous/mcp"], "env": {} }`
Any MCP client	its `mcpServers` block

# Codex. ~/.codex/config.toml
[mcp_servers.nous]
command = "npx"
args = ["-y", "@opennous/mcp"]

No API key is needed once you have run login. The server reads your key from ~/.nous/config.json. To pin a specific key instead, add "env": { "NOUS_API_KEY": "..." } with a key from Settings → API Keys.

Hosted server (remote HTTP)

For clients that cannot run a local process, like n8n cloud. Create an API key at Settings → API Keys, then add Nous with the URL and a Bearer token.

{
  "mcpServers": {
    "nous": {
      "url": "https://mcp.opennous.cloud/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

n8n cloud. MCP Client Tool node, Transport HTTP Streamable, Endpoint https://mcp.opennous.cloud/mcp, Auth Bearer.
Claude Code. claude mcp add --transport http nous https://mcp.opennous.cloud/mcp --header "Authorization: Bearer YOUR_API_KEY"
Hermes. In MCP settings set the name to nous, then paste the inner object { "url": "https://mcp.opennous.cloud/mcp", "headers": { "Authorization": "Bearer YOUR_API_KEY" } }.

The tools, by verb

Every tool is one of five verbs. Pick the verb, then the tool.

Verb	When to reach for it	Tools
Retrieve	pull what we already know	`get_context` · `get_account` · `query` · `search_notes` · `get_playbook`
Learn	what calls & replies taught us	`get_insights`
Aggregate	a metric across the whole workspace	`campaign_performance` · `pipeline_intelligence`
Act	what to work next	`attention` · `get_action_items`
Write	log what happened	`record` · `save_note` · `verify`

Retrieve — pull what we already know

get_context

Engineered context for a task about one entity. The headline call before outreach or a meeting.

get_account

The full account, every person and claim with epistemics plus the recent timeline.

query

Retrieve and summarise a corpus across many people — replies, meetings, the facts corpus.

search_notes

Semantic search across saved notes and documents.

get_playbook

Read the user’s own rules (voice, outreach, ICP, positioning) before you act — never query for this.

Learn — what buyers taught us

get_insights

What calls and replies taught us about our product, positioning, market, and buyer — grounded in verbatim quotes. The mirror of the playbook.

Aggregate — a metric across every account

campaign_performance

Outbound feedback loop — reply sentiment, conversions, and the ICP tier of who replied, by campaign and variant. Which copy to scale vs cut.

pipeline_intelligence

How the pipeline converts — stage distribution, stage-to-stage conversion %, median time-to-client and time-in-stage. Where deals leak and how long they take.

Act — what to work next

attention

A ranked to-do list — hot-intent accounts to reach out to now, accounts gone quiet, open commitments, upcoming meetings.

get_action_items

Open commitments grouped by account. What you owe them and what they owe you.

Write — log what happened

record

Record what happened. You observe, Nous derives the claims.

save_note

Keep a meeting brief, transcript, or note on a contact’s record — append-only and dated.

verify

Re-check a specific fact before you act on a stale-looking claim.

Use cases — try these prompts

Four things teams ask Nous every day. Each maps to the verbs above.

1. Account research — get_context / get_account

“Give me the full state of [account] — every person involved, their role, where each stands, and the open objections.”
“I’ve got a meeting with [account] — prep me and tell me what to focus on.”
“Write a contextual follow-up email to [account] based on our previous touchpoints.”

2. Pipeline patterns — pipeline_intelligence / attention / query

“Which accounts have gone quiet that I should re-engage?”
“Report the last 7 days of pipeline activity and what we should improve next week.”
“How long does it take us to follow up on positive replies, and where do deals stall?”

3. Feedback loops — campaign_performance / query

“Which campaign performs best and why, and what should we double down on?”
“What objections and pains show up across everyone who replied to our cold emails?”

4. Insights — get_insights

“What have our calls taught us about what to build next?”
“What language lands with buyers and what confuses them?”
“Who actually buys, and what’s the pain that drives the purchase?”

Setup — the agent operates the workspace

Nous is operated by the agent, not clicked through a UI. Call get_workspace_status first; it returns a ranked next-steps list.

get_workspace_status

The whole setup state plus ranked next steps. The first call in a session.

set_workspace_profile

Agent-driven onboarding. Set name, website, business type, and ICP.

build_icp_model

Build the ICP scoring model from the synced GTM context.

train_icp_model

Train the ICP model on real closed-won and closed-lost deals.

sync_icp

Sync the user’s ICP and context files into the graph (file to graph).

export_icp_model

Get the learned ICP model to write back into the ICP file (graph to file).

sync_playbook

Push an edited playbook file into Nous so every agent obeys the same rules.

connect_integration

Connect a key-based integration (Apollo, Prospeo, HubSpot token, and more).

Self-hosted vs Cloud. Lead Lists, CRM Sync, Triggers, and Reports are Cloud-only backend features (not agent tools). Everything else, including the ICP scoring model, is open on a self-hosted instance.

The daily loop

Start a task. Call get_context with the right intent for a ranked, budget-fit context block.
Need the user’s rules. Call get_playbook (voice, outreach, icp, positioning). Never use query or get_account for this.
Something happened. record it. Nous derives the claims.
Made something durable. save_note the brief or transcript onto the contact.
End of session. Call attention for what to work next.

record versus save_note. record logs that something happened (an email, a meeting) and Nous derives claims from it. save_note keeps the document itself (a brief, a transcript), append-only and dated. Use both.

Environment

Variable	Required	Default	Purpose
`NOUS_API_KEY`	if not signed in via CLI	none	Workspace key. Create at Settings, API Keys.
`NOUS_API_URL`	no	`https://api.opennous.cloud`	Override for self-hosted.

Source lives at apps/mcp. Each release ships to npm as @opennous/mcp@<semver>. Pin a version for reproducible installs with npx -y @opennous/mcp@0.43.0.

​Fastest setup

​Connect any client

​Hosted server (remote HTTP)

​The tools, by verb

​Retrieve — pull what we already know

get_context

get_account

query

search_notes

get_playbook

​Learn — what buyers taught us

get_insights

​Aggregate — a metric across every account

campaign_performance

pipeline_intelligence

​Act — what to work next

attention

get_action_items

​Write — log what happened

record

save_note

verify

​Use cases — try these prompts

​Setup — the agent operates the workspace

get_workspace_status

set_workspace_profile

build_icp_model

train_icp_model

sync_icp

export_icp_model

sync_playbook

connect_integration

​The daily loop

​Environment

Fastest setup

Connect any client

Hosted server (remote HTTP)

The tools, by verb

Retrieve — pull what we already know

Learn — what buyers taught us

Aggregate — a metric across every account

Act — what to work next

Write — log what happened

Use cases — try these prompts

Setup — the agent operates the workspace

The daily loop

Environment