MCP server + API for AI agents

Let your AI agent run Juniors AI.

Juniors AI ships an agent-facing layer so the AI tools your team already works in can operate the platform directly. Create a scoped organization API key, drop one config block into Cursor, Claude Desktop, Claude Code, Windsurf, or any other MCP client, and your agent gets 33 tools across all six products - reading fleet health, querying production data in plain English, running SEO skills, reconciling settlements, and auditing calls. Every tool is generated from a single typed manifest that also produces the public OpenAPI spec, so the MCP tools and the documented REST API never drift apart. Keys are scoped per product and per access level, so an agent only ever sees the tools you granted it.

33tools
6products
28read-only
5writes

Get Access Setup in 3 steps

mcp.json

{
  "mcpServers": {
    "juniors-ai": {
      "command": "npx",
      "args": ["-y", "@juniors-ai/mcp"],
      "env": {
        "JUNIORS_API_KEY": "jun_live_..."
      }
    }
  }
}

Cursor
Claude Desktop
Claude Code
Windsurf
Cline
Any MCP client

Setup

Connected in under five minutes

Step 01
Create a scoped API key
In the Juniors AI dashboard, open Admin -> API Keys and create a key. Pick the products and the access level it needs. The key is shown once, so copy it then.
Step 02
Add the server to your client
Drop one block into your MCP client config. There is nothing to install and no SDK to wire up - npx fetches the server on demand.
Step 03
Restart and start asking
On reconnect the server checks your key and registers only the tools your scopes allow, so the tool list is exactly what the key can do.

mcp.json

{
  "mcpServers": {
    "juniors-ai": {
      "command": "npx",
      "args": ["-y", "@juniors-ai/mcp"],
      "env": {
        "JUNIORS_API_KEY": "jun_live_..."
      }
    }
  }
}

Set JUNIORS_API_URL as well if you need to point the server at a staging or self-hosted deployment. The package is published as @juniors-ai/mcp.

How it works

Built for agents, scoped for humans

One config block, 33 tools

Add the server to your MCP client with npx and an API key. On reconnect your agent has tools for Concierge, Data Sensei, Search Sensei, Recon, Sonar, and Sentinel - no SDK to install, no glue code to write.

Scoped keys, least privilege

Every key is scoped per product and per access level. Scopes are <product>:read or <product>:write, and write implies read. A key can never exceed the permissions of the member who created it.

Only the tools you granted

The server calls whoami on startup and registers only the tools your key is scoped for. A Sonar-only key shows a short, relevant tool list instead of dozens of tools that would just fail with a permission error.

Read-heavy by design

28 of the 33 tools are read-only. The five writes are deliberate and narrow: start a chat, ask a question, run an SEO skill, trigger a reconciliation, ask a call a question. Nothing destructive is exposed.

A real OpenAPI spec

The same typed manifest that generates the MCP tools generates the OpenAPI document served at /api/openapi.json. Point any HTTP client, code generator, or custom agent framework at it - the spec can never drift from the tools.

Your data stays yours

Agent traffic follows the same bring-your-own-infrastructure rules as the dashboard. Tools read from your database, your storage, and your LLM keys. Juniors AI never gains a copy of the data an agent touches.

Auditable and revocable

Keys are stored as SHA-256 hashes, shown exactly once at creation, rate limited per key, and revocable instantly from the dashboard. Last-used timestamps tell you which keys are actually live.

Works with the whole workspace

One key can span every product your member account can reach, so a single agent session can pull a Sentinel incident, check the Recon run behind it, and read the Concierge tickets customers filed about it.

Tool reference

33 tools across all six products

Scopes are <product>:read or <product>:write, and write implies read. juniors_whoami needs no scope and is always available, so an agent can always discover what it is allowed to do.

Concierge

AI Customer Support

concierge

readconcierge_list_conversations
readconcierge_get_conversation
readconcierge_analytics_overview

“List every open WhatsApp conversation, then summarise the three complaint themes that come up most often.”

Data Sensei

Natural-language Analytics

data-sensei

readdata_sensei_list_sources
readdata_sensei_list_chats
readdata_sensei_get_chat
writedata_sensei_create_chat
writedata_sensei_send_message

“Open a chat on my Postgres source, ask for weekly active users over the last 8 weeks, and show me the SQL you ran.”

Recon

Financial Reconciliation

recon

readrecon_list_sources
readrecon_list_comparisons
readrecon_list_runs
readrecon_get_run
readrecon_get_run_rows
writerecon_trigger_run

“Reconcile yesterday's settlement comparison, then list every row that landed in the amount-mismatch bucket.”

Search Sensei

AI SEO + GEO Agents

search-sensei

readsearch_sensei_list_skills
readsearch_sensei_list_runs
readsearch_sensei_get_run
readsearch_sensei_performance
writesearch_sensei_run_skill

“Run keyword research for example.com, then generate a content brief for the best opportunity you find.”

Sonar

AI Call Auditing

sonar

readsonar_list_calls
readsonar_get_call
readsonar_overview
readsonar_hotspots
readsonar_voc
writesonar_ask_call

“Find this week's calls flagged for attention, then explain what the agent did wrong on the worst one.”

Sentinel

Autonomous DevOps

sentinel

readsentinel_overview
readsentinel_monitoring
readsentinel_alarms
readsentinel_cost
readsentinel_security
readsentinel_activity
readsentinel_cicd_deployments

“Check fleet health, then summarise every open incident with its root cause and what you would do about it.”

Not using MCP?

It is a documented REST API too

Every tool is generated from one typed manifest, and that same manifest generates the OpenAPI 3 document. Point any HTTP client, code generator, or custom agent framework at the spec - it can never drift from the tools the MCP server exposes.

View the OpenAPI spec

curl

curl https://app.juniors.ai/api/agent/whoami \
  -H "Authorization: Bearer jun_live_..."

# Returns the org, the member, and every
# product scope the key is allowed to use.

MCP FAQ

MCP server - common questions

What is the Juniors AI MCP server?

It is a Model Context Protocol server, published as @juniors-ai/mcp, that exposes Juniors AI as tools any MCP client can call. Add it to Cursor, Claude Desktop, Claude Code, Windsurf, Cline, or any other MCP client with a scoped organization API key, and your agent gets 33 tools spanning all six products: Concierge, Data Sensei, Search Sensei, Recon, Sonar, and Sentinel.

How do I install the Juniors AI MCP server?

Which MCP clients are supported?

How do API key scopes work?

What can an agent actually do?

Is there a REST API and an OpenAPI spec?

Is the agent surface secure?

Does using the MCP server cost extra?

Give your agent the keys

Get early access to the full Juniors AI workspace, MCP server included at no extra cost.

Get Access