Crews that share institutional knowledge

Because even the best crews need a company wiki

The Problem

CrewAI is brilliant at orchestrating agents. Your researcher finds information, your analyst processes it, your writer creates content. Beautiful coordination.

Then you end the session.

Tomorrow, your crew assembles again—a team of experts who've never met. The researcher doesn't remember what they found. The analyst has no notes. The writer stares at a blank page wondering what project this even is.

It's like running a company where everyone gets memory-wiped at 5 PM. Great teamwork, zero institutional knowledge.

Your crew has impressive job titles and absolutely no context about their actual job.

How Stompy Helps

Stompy becomes your crew's institutional memory—the company wiki that actually gets maintained.

When your agents work, they build shared context:

- Research findings persist: Your researcher's discoveries are available to the whole crew, forever - Decisions compound: Each session builds on previous conclusions - Roles gain depth: Your analyst remembers their analytical framework - Handoffs improve: Agents inherit context from their predecessors

It's the difference between a crew that works together and a crew that works together over time.

Your agents don't just coordinate within sessions—they build on everything they've ever learned together.

Integration Walkthrough

Add Stompy to your agents via mcps field

CrewAI has native MCP support. Add Stompy using SSE transport with bearer auth.

from crewai import Agent, Task, Crew
from crewai.mcp import MCPServerSSE
import os

# Create agent with Stompy memory via SSE
researcher = Agent(
    role="Research Analyst",
    goal="Research and remember findings across sessions",
    backstory="Expert researcher with persistent memory",
    mcps=[
        MCPServerSSE(
            url="https://mcp.stompy.ai/sse",
            headers={"Authorization": f"Bearer {os.environ['STOMPY_TOKEN']}"}
        )
    ]
)

Researcher saves findings to shared memory

When agents discover important information, they save it with lock_context. All crew members can access it.

# Week 1: Research crew does competitive analysis
research_task = Task(
    description="Analyze competitor pricing strategies",
    expected_output="Detailed pricing analysis",
    agent=researcher
)

crew = Crew(agents=[researcher], tasks=[research_task])
crew.kickoff()

# Researcher saves findings to shared memory:
# lock_context(topic="competitor_pricing_q4",
#   content="CompetitorA: $99/mo starter, $299 pro.
#            CompetitorB: usage-based at $0.01/request.
#            Market average: $150/mo for comparable features.",
#   tags="pricing,competitors,market-research")
# → Creates v1.0

Strategy crew retrieves research later

Weeks later, different crew members retrieve the research using context_search or recall_context.

# Week 3: Strategy crew needs that research
strategist = Agent(
    role="Pricing Strategist",
    goal="Develop competitive pricing strategy",
    mcps=[MCPServerSSE(
        url="https://mcp.stompy.ai/sse",
        headers={"Authorization": f"Bearer {os.environ['STOMPY_TOKEN']}"}
    )]
)

strategy_task = Task(
    description="Develop pricing strategy based on our research",
    expected_output="Pricing recommendation with rationale",
    agent=strategist
)

crew = Crew(agents=[strategist], tasks=[strategy_task])
crew.kickoff()

# Strategist calls: context_search("competitor pricing")
# Strategist: "Based on our Q4 research showing market
#   average at $150/mo, I recommend $129/mo to undercut
#   while maintaining margins..."

Crew knowledge compounds over time

Each session adds to institutional knowledge. Research builds on research. Decisions reference past decisions.

# Month later: Update pricing research
crew.kickoff(inputs={"task": "Update Q1 competitor analysis"})

# Researcher recalls previous, adds new findings:
# lock_context(topic="competitor_pricing_q4",
#   content="Q1 Update: CompetitorA raised to $119/mo.
#            CompetitorB added enterprise tier at $500/mo.
#            Market shifting toward value-based pricing...")
# → Creates v1.1, preserving v1.0 history