Integrating MCP: The Start of Agents That Feel Like Teammates

Common Team

• 3 min read
Integrating MCP: The Start of Agents That Feel Like Teammates
Integrating MCP: The Start of Agents That Feel Like Teammates

Today we’re introducing MCP (Model Context Protocol) slash-mentions in Common, so people and agents can finally work side-by-side in the same thread.

We’ve learned two things: crypto UX gets buried in tabs, and AI agents underperform when they lack context. MCP gives us a shared language so Common can treat tools and agents like first-class teammates, right inside threads.

TL;DR

  • Type / to summon trusted tools and agents inside a conversation.
  • Mentions expand into real tool calls with scoped, temporary permissions.
  • Results land in-thread where your team can reuse them.
  • Rolling out in beta with a focused set of integrations; expanding based on feedback.

Try it today

  1. In the Admin Panel in any community, open Admin Capabilities → Integrations → MCP
  2. Enable an integration (e.g., Google Sheets) and complete OAuth
    • We currently support any transaction available through Klavis
  3. Ship the work—in one place, with context
    • Everything stays in context so teammates can follow, reuse, or continue the thread.

In any thread, type / and pick your agent

hey [/google_sheets], list our quarterly budget sheets

MCP slash-mentions turn chat into a command line for your team.

Summon an agent, keep the context, and move the work forward.

Simple actions: where your community already works

1) Spec → Ship (Product & Eng)

Outcome: turn ideas into PRs and payouts without leaving the thread.

How it looks:

/research_bot landscape on "prediction markets for collaboration"
/github open PR "bounty splits in thread composer"
/mint bounty "QA checklist" 100 C split @alex 50% @sam 50%
/vote approve merge when checks pass

2) Finance & Ops Loop (Sheets/Notion)

Outcome: close the budget loop in-thread with auditable changes.

How it looks:

/google_sheets pull "Q4 Budget"
/ops_bot flag over-budget line items > 10%
/google_sheets update "Q4 Budget" B12=24500 note="Approved by FP&A"
/vote approve reallocation

3) Trading & Research (Investing)

Outcome: move from research to action with clear guardrails.

How it looks:

/research_bot summarize last 5 onchain events for $TOKEN
/dune run dashboard 12345 filter: token="$TOKEN"
/vote queue "deploy strategy v2" with max gas $50 → hold for approval

4) Governance & Treasury

Outcome: proposals, approvals, and safe disbursements without key sprawl.

How it looks:

/governance draft proposal "Fund community translators" 2,500 C
/tests simulate "net effect on runway"
/approvals route @council → 2-of-3 multisig
/payments execute when threshold met

From here: proactive & background agents

MCP slash-mentions make agents respond inside the thread. As we gather feedback, we want to enable these agents to act in the background and proactively aid the community and workspace—always with clear controls.

What “proactive” means:

  • Event-based triggers: new PR, budget variance > X%, on-chain signal, contest end time, whale trade, KPI breach.
  • Time-based routines: daily standup digest, weekly budget roll-up, monthly retro of bounties vs shipped work.
  • Goal-based nudges: if a space OKR is off-track, pre-draft bounties, surface owners, or suggest splits.

How it shows up:

  • Agents post summaries, suggested actions, and pre-filled commands into the thread for one-tap approval.
  • For high-trust flows, agents can auto-execute within scoped budgets/limits and post receipts.

Controls & safety:

  • Budgets & rate limits per space/agent; allowlists for what can be automated.
  • Approval policies (e.g., 2-of-N for treasury) and tests/simulations before funds move.
  • Memory scopes & retention windows so background learning stays privacy-bounded.
  • Audit trail: every background action posts a signed receipt back to the thread.

Example (background pilot):

  • Every morning, /ops_bot checks “Q4 Budget.” If any line drifts >10% week-over-week, it posts a digest + /vote reallocate draft.
  • After a PR merges with “bounty” label, /payments suggests the split from the thread context and awaits one-tap approval.
  • Before a contest closes, /community_bot pings judges with a ranked shortlist and drafts payout transactions.

What’s next (light roadmap)

MCP slash-mentions are a first step in a broader arc, as we discussed in our roadmap.

Our long term goal, as we gather feedback, we want to enable these agents to act in the background and proactively aid the community and workspace—nudging critical follow-ups, opening bounties, and drafting updates without being asked (always with clear controls and approvals).