Skip to main content

Set up and use the Modjo MCP

The Modjo Model Context Protocol (MCP) server lets your AI assistant β€” Claude, ChatGPT, Cursor, Dust, n8n, and any other MCP-compatible client β€” read and analyze your Modjo workspace data: calls, deals, accounts, contacts, transcripts, and AI agents. You ask in plain language, Modjo answers from your actual data.

Feel free to check our clients use cases of Modjo's MCP to get inspiration.

How to connect to Modjo MCP?

We support two authentication methods. The choice matters for security and team rollout, so read this before you start.

OAuth 2.0 (recommended β€” use this by default)

Each user signs in with their own Modjo account.

Use OAuth for the entire team, including managers and admins, in 99% of cases.

Bearer token via HTTP header (admin / power-user only)

A workspace API key passed as an Authorization header. The token is organization-scoped β€” whoever holds it can read every call, every deal, and every account across the workspace, regardless of their role in Modjo.

This is the right tool when you genuinely need workspace-wide reach: building n8n pipelines, programmatic automations, or scripts owned by an admin. It is the wrong tool for individual reps. If a rep can use OAuth, they should.

OAuth

Bearer token

Who it's for

All users β€” reps, CSMs, managers

Admins, ops, automation owners

Setup

Click "Connect", sign in

Generate API key, paste header

Best for

Daily use in Claude, ChatGPT, Cursor

n8n workflows, scripts, server-side jobs

Token rotation

Automatic

Manual


Quick start with OAuth (most clients)

Step 1: Add custom connector to your workspace

This step should be carried on by the admin within your organization for Claude / ChatGPT / Dust…

Here are the details for each client:

Claude (claude.ai web, Claude apps, Claude Code)

  1. Go to Customize β†’ Connectors β†’ Add custom connector

  2. Name: Modjo

  3. URL: https://api.mcp.modjo.ai/v1/mcp

  4. Click Connect, sign in with Modjo, approve access

The Modjo tools become available to Claude immediately. You can scope them per-project in Claude Projects.

ChatGPT

  1. Open Settings β†’ Connectors β†’ New connector (requires ChatGPT Plus or higher, depending on rollout)

  2. URL: https://api.mcp.modjo.ai/v1/mcp

  3. Sign in with Modjo when prompted

Cursor

  1. Settings β†’ MCP β†’ Add new server

  2. Paste https://api.mcp.modjo.ai/v1/mcp

  3. Approve OAuth in the browser when Cursor prompts

Dust

Dust supports remote MCP servers natively.

  1. Spaces β†’ Tools β†’ Add Tools β†’ Add MCP Server

  2. URL: https://api.mcp.modjo.ai/v1/mcp

  3. Choose OAuth as the auth method, complete the sign-in

  4. Assign Modjo tools to agents in the Agent Builder

For automation-style Dust agents where a single API key per workspace is acceptable, the Bearer Token option is also supported (see Bearer token setup below).

Other MCP-compatible clients

Any client supporting the MCP OAuth 2.0 spec (Authorization Code + PKCE with protected-resource metadata discovery) will work. Point it at https://api.mcp.modjo.ai/v1/mcp and let it handle the rest β€” the server publishes its OAuth metadata at the standard .well-known paths.

Step 2: Connect as a user

Once the custom connector has been added to your workspace all you need to do is go to your workspace’s connectors (path depends on which client you use - see above), select Modjo and sign in! As simple as that.

Good to know: If you had already setup the Modjo MCP with the bearer token, we recommend you to switch to the Oauth method for simpler usage by your whole team.
Also, don't forget to change the MCP source connected to your agents if you had already setup agents using the Modjo MCP!


Bearer token setup (admins & automations)

Use this only when OAuth isn't a fit β€” typically n8n workflows, scheduled scripts, server-to-server pipelines, or one-off admin queries that need workspace-wide reach.

1. Get your API key

  1. Generate a key and store it somewhere safe β€” you won't see it again

2. Configure your client

The bearer token uses a slightly different endpoint that accepts the Authorization header directly:

<https://api.mcp.modjo.ai/v1/mcp>

Pass your key as Authorization: Bearer <your-api-key>.

n8n (version 1.88.0+)

n8n natively supports MCP since v1.88.0 β€” no community node, no extra install.

Mode A β€” With an AI Agent (autonomous tool calls):

  1. Add an AI Agent node with your LLM credentials

  2. Add an MCP Client Tool node, connect it to the Agent's Tools input

  3. Configure the MCP Client Tool:

    • MCP Endpoint URL: https://api.mcp.modjo.ai/v1/mcp

    • Authentication: Header Auth

    • Header name: Authorization

    • Header value: Bearer your-modjo-api-key-here

  4. Under Tools to Include, choose All (or Selected for specific tools)

Mode B β€” Standalone (deterministic tool calls):

  1. Add an MCP Client node

  2. Configure:

    • Server Transport: Streamable HTTP

    • MCP Endpoint URL: https://api.mcp.modjo.ai/v1/mcp

    • Same Header Auth as Mode A

  3. Select the Tool to execute from the dropdown

  4. Map input parameters

Make / Zapier / Codex CLI / Gemini CLI / any HTTP MCP client

Same pattern everywhere:

{   "mcpServers": {     "modjo": {       "type": "http",       "url": "<https://api.mcp.modjo.ai/v1/mcp>",       "headers": {         "Authorization": "Bearer your-modjo-api-key-here"       }     }   } }

Security reminders for bearer tokens

  • Don't commit API keys to git. Use environment variables or your secrets manager.

  • Don't share keys between users. One key per automation or admin, rotated regularly.

  • A bearer key is workspace-wide. Anyone holding it can see every call, every deal, every account, including data from teams they don't normally have access to. Treat it like an admin password.

  • Prefer OAuth for individuals. Even admins doing daily analysis in Claude or ChatGPT should sign in with OAuth β€” keep the bearer token for systems, not seats.


Available tools

Twelve tools, three categories. Modjo agents are the most powerful β€” they produce structured analysis instead of raw data.

Search & retrieval

  • get_accounts β€” list/search accounts (CRM ID, link, etc.)

  • get_deals β€” list/search deals (amount, dates, status, source, loss reason)

  • get_calls β€” list/search calls (filter by deal, account, contact, date, user)

  • get_contacts β€” list/search contacts (CRM person ID, email, phone, title)

  • get_users β€” list/search Modjo users (role, department)

  • get_emails β€” list/retrieve emails with content

  • semantic_search_activities - retrieve activities (calls or emails) about a specific topic (e.g. competition, objection, pricing, etc.)

Content

  • get_transcript β€” full transcript of one or more calls, with timestamps and speakers

AI analysis (agent-first)

  • ask_anything_on_call β€” AI question scoped to a single call

  • ask_anything_on_deal β€” AI question scoped to a deal

  • ask_anything_on_account β€” AI question scoped to an account

Agent catalog

  • get_agents β€” list available agents (Modjo built-in or custom)

πŸ’‘ Sometimes AI tools need manual refresh to acknowledge we added new tools to the MCP. If you notice that one is missing, ask your admin to refresh the connection with the MCP in the AI tool settings.


Using Modjo agents through the MCP

Instead of open-ended questions, point Modjo's specialized agents at your data. You get the same structured output every time, which makes results comparable across reps and deals.

Built-in agents

Agent

UUID

Best for

CallSummary

741e9ffc-87be-4bca-bb0d-f167be8b963e

CRM-ready call recap

NextStepper

09715241-0cdd-44c9-a386-92a1340bdf4a

Actions and commitments (us vs them)

MeetingPrepper

c0a76fd7-3f56-4a03-b2e7-7765da10c457

Pre-meeting brief + questions to ask

CallQualifier

3a7753e1-d21e-4cb7-a990-7820291274cd

Structured qualification + gaps

DealBriefing

1204e84f-6edd-4782-bbdf-e5e070b400cf

Executive deal summary + risks

EmailFollowUp

b2a9ae1b-2026-4dfa-9c67-101733a87a04

Follow-up email draft

Example: pull next steps from a call

  1. Find the call with get_calls

  2. Ask Claude: "Use the NextStepper agent on call [ID] to extract every commitment. Separate our tasks from theirs, include dates."

  3. The assistant calls ask_anything_on_call with agentUuid: "09715241-0cdd-44c9-a386-92a1340bdf4a"

  4. You get structured client-side and team-side actions with timestamps

Example: executive deal briefing

  1. Get the deal CRM ID with get_deals

  2. Ask: "Use DealBriefing on deal [CRM_ID] for an actionable briefing β€” status, stakeholders, signals from recent calls, risks, expected decisions, close plan."

  3. You get a document ready for forecast review or deal desk


Other relevant information

Best practices

  • Standardize questions per agent. Same wording β†’ comparable outputs across the team.

  • Anchor the time scope. "Over the last 30 days" or "since the last call" β€” otherwise the agent crawls everything.

  • Ask for evidence. Add "cite what was said and by whom" to force timestamped quotes.

  • Use pagination. Search tools cap at 50 results per page; iterate with nextCursor when you need the full set.

  • Combine filters. get_calls accepts deal + dateRange + userIds simultaneously β€” narrow before you analyze.

  • Document your agent UUIDs. A team-internal cheat sheet of agent β†’ use case speeds adoption.


Limits & constraints

  • Date format: YYYY-MM-DD for any dateRange filter. Anything else throws.

  • No stats tool. Need call counts or total duration? Pull via get_calls, compute client-side.

  • Pagination required. Max 50 results per page across all search tools.

  • UTF-8 in searches. Use Γ©, Γ¨, Γ  literally β€” not \\uXXXX escapes.

  • AI latency. ask_anything_* can take several seconds on long transcripts. Some clients enforce a 60-second tool timeout β€” keep that in mind for deep account-level queries.

  • Scope: the MCP can retrieve data from any call and doesn't yet abide by call access rights you may have implemented in the app (it will soon be the case)

  • Semantic search tool: it will only consider the top 100 most relevant activities. If you want an exhaustive result, use a time filter for more relevant results (e.g. what were the top objections raised last week?)


Troubleshooting

OAuth sign-in loops or fails

Make sure you're already signed in at app.modjo.ai in the same browser before starting the OAuth flow. Clear cookies for modjo.ai if it keeps looping.

Tools don't appear after connecting

Wait 30 seconds and refresh. If they still don't show, check that the URL is exactly https://api.mcp.modjo.ai/v1/mcp (no trailing slash, no typos) and that your client supports remote MCP with OAuth.

401 / 403 errors with bearer token

Check the key is correct and the header is exactly Authorization: Bearer <key> (with the space). Test with curl:

curl -H "Authorization: Bearer your-key" <https://api.mcp.modjo.ai/v1/mcp>

ask_anything_on_deal or _account times out

These crawl a lot of data. Two causes:

  • The client enforces a 60-second tool timeout

  • The deal/account has hundreds of calls

Narrow the scope (date range, specific call IDs) or use the call-level variant.

Tools work but data feels stale

MCP queries are near real-time. There's typically a few-second delay between a call ending and its availability via MCP β€” wait a minute and retry.

Did this answer your question?