Connect your AI assistant
Connect Claude, ChatGPT, Cursor, or any MCP client to your SuperOrgs organization. About a minute, no API keys.
MCP endpoint
https://app.superorgs.com/mcpAdd the connector
Pick your client and follow the steps. The endpoint is the same everywhere.
- 1
Open connector settings
In Claude, go to Settings, then Connectors. - 2
Add a custom connector
Choose Add custom connector and paste the endpoint.https://app.superorgs.com/mcp - 3
Connect and pick your org
Click Connect, sign in to SuperOrgs, and choose your organization on the consent screen. - 4
Ask something
Try “What did my agents do today?” Claude will call the right tools.
What you can ask
Ask in plain language. SuperOrgs picks the right tools and answers from your real workforce.
Tools reference
Every tool the server exposes, grouped by what it answers. Names are stable; descriptions come straight from the product.
| Tool | What it returns |
|---|---|
| Activity | |
activity_org_feed | Recent runs across every agent in the org, newest first, each with its Orion summary, status, error message, event count, and cost. Orion's own runs report cost as null: that run spend is SuperOrgs' vendor cost, not the org's. |
activity_stats | Org-wide activity headline for the last 7 days: runs, active agents, errors, spend. |
activity_agent_runs | One agent's recent runs (including its linked deployments), newest first, with Orion summaries and cost. Orion's own runs report cost as null: that run spend is SuperOrgs' vendor cost, not the org's. |
activity_run_events | The full event timeline for one run (messages, tool calls, cost), oldest first. Orion's own runs carry no cost events. |
activity_summarize_agent | A digest of what one agent recently did: its latest runs with Orion summaries plus 24h stats. Use this to answer "what did agent X do". |
| Workforce | |
agents_list | Every agent in the org: name, type, platform, status, owner, department, monthly cost. |
agents_get | One agent by id, with full detail. |
org_chart_departments | Departments with their human headcount, agent count, human-to-agent ratio, and monthly AI spend. |
workforce_stats | High-level organization statistics: employee count, agent counts by status, total workforce, monthly AI spend, and the agent-to-employee ratio. Fields the caller lacks permission to read are null. |
| People | |
employees_list | Active employees (capped at 50): id, name, title, email, department. Optionally filter by department name (case-insensitive exact match). |
employee_search | Find employees by a case-insensitive substring of their display name. |
| Costs | |
costs_summary | Monthly AI spend, human-equivalent cost, savings, and ROI across the org. |
cost_breakdown | Monthly AI agent cost grouped by department, platform, or agent type, sorted highest to lowest. |
cost_top_agents | The biggest-spend agents over the window, each with its real run cost in cents and its share of the spend shown here. Spend metered against an agent that has since been removed cannot be named, so it is reported once as unattributedCents rather than as a row. |
cost_anomalies | Agents whose run cost in the trailing window is at least 2x the equal-length window before it (a rate comparison, not a projection). New spenders with no prior baseline are flagged. |
| Health | |
agents_health | Per-agent run count, error count and rate, average duration, and idle time over the window, ranked worst-first to surface failing or stale agents. |
agent_errors_recent | The most recent failed runs across the org, newest first, each with the agent name, error message, cost, and time. Orion's own runs report cost as null: that run spend is SuperOrgs' vendor cost, not the org's. |
run_forensics | One run's full story for debugging why it failed or what it did: the run header (status, error, cost, tokens, duration, summary) plus its complete event timeline. Orion's own runs report cost as null: that run spend is SuperOrgs' vendor cost, not the org's. |
| Discovery | |
agent_search | Find agents by a substring of their name, title, description, or department. |
automation_gaps | Departments with active employees but little to no agent coverage, ranked by the widest gap, to show where deploying agents would help most. |
| Pods | |
pods_capacity | Each pod's estimated build capacity, actual work underway (In Progress and Onboarding), and queued agents with projected start dates. Underway describes lifecycle, not a live runtime session. Work underway remains underway when staffing is missing or capacity is exceeded. Slots estimate concurrent builds, never people; builderCount and expertCount are headcounts. Slots and dates are planning estimates, not measured availability or delivery commitments. startsThisPeriod means unscheduled work could start now, not that it has started. Saved plannedStartsOn dates take precedence over automatic estimates. plannedThisPeriod counts queued commitments separately from underway work. Use pod_insights for the full saved plan. |
pods_suggestions | The open pod suggestions Orion has proposed for this org: each with its suggestion id, proposed name and mission, the reasoning, and the suggested members and agents with names and per-pick why lines. Use before accepting or dismissing a suggestion, or for questions like "what pods are being suggested and why". |
pod_insights | Read a pod's evidence-based attention items with source links, recorded progress over the last 7 days, and saved current/next build plan. Missing stage history excludes time-based alerts. Counts reflect recorded activity, not all historical work. Plans retain their dates and never change agent status. |
slack_channel_read | Read the pod's bound Slack channel one page at a time, newest first. Returns messages with author names, thread markers (threadTs + replyCount on thread parents), and a nextCursor when older messages exist: pass it back to walk further into history, and pass threadTs to read a thread's replies. Be honest about coverage: say how far back you actually read. Use for questions about what the pod is discussing or deciding. |
teams_channel_read | Read the pod's bound Microsoft Teams channel one page at a time, newest first. Every page already carries each thread's replies inline, marked with inThread, so there is no separate thread read to make. Author names are resolved against the org directory, and a nextCursor comes back when older messages exist: pass it back to walk further into history. The SuperOrgs app's own posts are not returned, only what people said. Be honest about coverage: say how far back you actually read. Use for questions about what the pod is discussing or deciding. |
| Ideas | |
idea_inbox_list | Recent ideas on the Ideas board: name, status, where each came from, and whether Orion has already suggested next steps for it. Newest first. Each idea may carry Orion's best-fit shape (agent, task, or general). |
idea_sources_list | The connected sources Orion may scan for ideas, each with its approved scope (Slack channels on the analysis allowlist; projects in the saved scope) and the consent note describing what a scan reads. A scan may only cover what appears here. |
idea_scan_status | The latest ideas scan for a provider: its status, what it covered, and how many detections were staged on the Detected tab. "stopped" means a person stopped the scan, so its counts cover only what ran before that. Check this before reporting any scan as finished; never claim results you have not read from here. |
| Planning | |
agent_priority_scores_list | Every suggestible agent with its AI priority score: the effective score (0-100, an override when set), whether that score is a human override, whether the org rubric has changed since it was generated, and the rationale behind it. Sorted highest score first; unscored agents last. |
| Company research | |
research_list | The index of the company research: every concept with its path, type, title, and one-line description. Concepts cover the company, departments, publicly named leaders, signals, and suggested agents with their receipts. Pull this first, then read specific concepts with research_read. |
research_read | One full research concept by path (from research_list): its type, title, sources, and complete body. Bodies carry inline source URLs next to each claim; cite them when answering from this. |
Two prompts ship with the tools: workforce_review for a weekly review and investigate_cost_spike for a spend spike. Three resources attach from your client's attachment menu: the org chart snapshot, an agent profile, and a run report.
Permissions and safety
- Your token is scoped to your organization. The org comes from the verified token, never from anything you type.
- Every tool runs the same permission checks as the app. Your role and owner scope are respected.
- Version 1 is read-only. Tools answer questions; actions come later.
Troubleshooting
- Tools list is empty or stale
- Remove the connector and add it again so your client refreshes the tool list.
- 401 or an auth loop
- Your session expired. Reconnect and sign in again.
- Connected but no data
- Check you picked the right organization on the consent screen.