Wired into every tool you already use.
Five day-one connectors. An MCP server so any AI client can tap into your company’s pulse. A public REST API and outbound webhooks for programmatic integration. Bring-your-own-LLM-key for procurement-friendly deployments. Federated insights without federated data.
Five tools cover 85% of where work lives.
We could ship 100 connectors. We chose 5: the ones modern teams actually live in. More land through Q3/Q4 based on demand, never to chase a feature-checklist.
Slack
Channels, threads, DMs · ACL-mirrored
GitHub
Issues, PRs, reviews, commit metadata
Notion
Pages + databases · workspace ACLs
Linear
Cycles, projects, issues, comments
Calendar
Events you organize / attend · OOO
Otter / Granola / Fireflies
Meeting transcripts → Decision Memory
Google Drive
Docs, Sheets, Slides · queued by demand
Confluence
Spaces, pages, comments
Jira
Issues, sprints, custom fields
Pulse in every AI tool you already use.
A standardized Model Context Protocol server at mcp.pulse.app. Any MCP client (Claude, Cursor, Cowork, ChatGPT) can connect via OAuth and query your company’s pulse. Read-only in v1, always permission-checked as the requesting user.
- search_company
Hybrid search across all entities · returns ranked results with citations
- get_decision
Fetch a decision by ID with full evidence trail
- list_recent_decisions
Last N decisions, filtered by team/topic/date
- find_expert
Given a topic or problem description, return ranked humans + availability
- get_feature
Full feature record: stages, owners, ship history, linked decisions
- list_similar_features
Given a one-paragraph description, return historically similar past features for Planner-style research
- get_person
Person profile + inferred skill graph + current availability
- get_pulse
User's current Pulse: briefing, diff, signals
- list_commitments
Open commitments by or to a given person, filterable by status
read-only todayWrite actions (create commitments, mark decisions resolved) are on the roadmap with explicit per-tool grants.
{
"mcpServers": {
"pulse": {
"url": "https://mcp.pulse.app",
"transport": "sse",
"auth": {
"type": "oauth",
"clientId": "$YOUR_CLIENT_ID"
}
}
}
}Pull, post, subscribe.
Six REST endpoints over decisions, features, commitments, people, topics, and Ask. Outbound webhooks for every entity event Pulse emits, signed with HMAC-SHA256 so your handler can trust the payload. Generate a workspace-scoped bearer token at /app/admin/api-keys; subscribe to events at /app/admin/webhooks.
- GET/v1/decisions
Filter by since · topic · cursor · limit
- GET/v1/features
Filter by stage · status (shipped / in-flight)
- GET/v1/commitments
Filter by owner · status
- GET/v1/people
List workspace members
- GET/v1/topics
List recurring topics
- POST/v1/ask
Body: { query, mode: "quick | deep | counter" }
- POST/v1/documents
Push a custom doc into Pulse · idempotent on retry · /docs/push-api
- POST/v1/graphql
Queryable knowledge graph · /docs/graphql
Responses are wrapped in { data, meta }; errors return { error } with a matching HTTP status. Cursor-paginated.
curl https://api.pulse.app/v1/decisions \ -H "Authorization: Bearer pk_live_xxxxxxxxxxxx"
Every entity event Pulse emits: decision.*, commitment.*, feature.*, agent_action.*, access_request.*, customer.churn_risk.
Delivery is at-least-once with exponential backoff (1m / 5m / 15m / 1h / 6h). Verify the X-Pulse-Signature header before trusting the payload; the verification snippet lives at the API docs.
Your model contracts. Your audit trail.
Pipe your own Anthropic, OpenAI, Azure, or Bedrock key. Pulse charges a flat per-seat fee with no LLM markup; your model bill goes directly to your provider. Procurement-friendly · cost-passthrough · audit-ready.
Anthropic
Claude Sonnet · Opus · Haiku
flat per-seat · no LLM markupOpenAI
GPT-4o · o-series · embeddings
flat per-seat · no LLM markupAzure OpenAI
Enterprise tenant · regional endpoints
flat per-seat · no LLM markupAWS Bedrock
Claude · Llama · provisioned throughput
flat per-seat · no LLM markupFederated insights, never federated data.
Pulse calibrates against an opt-in network of customers, 47at last count. Aggregate signals like “companies your size estimate features at 4w but ship at 7w” become first-class context. Your data never leaves your tenant.
Companies your size estimate features at 4w but ship at 7w on average
based on 1,240 features across 47 customers67% of decision-to-execution gaps close within 21 days of being surfaced
based on 3,820 decisions across 47 customersTeams with permission-aware AI see 2.4× the Ask Bar engagement
based on longitudinal · 12 quartersCompanies onboarding 3+ engineers simultaneously lose 1.5 quarters of velocity
based on 82 hiring wavesWhat stays true across every integration.
Permission-aware everywhere
MCP queries run as the user · same Cerbos filter as in-app · the AI client only sees what the user can see in source systems.
Read-only by default
MCP today is read-only. Write actions are coming with explicit per-tool grants. Pulse never moves data without authorization.
Bring your own keys
Anthropic, OpenAI, Azure, Bedrock: pipe your own LLM key. Cost passthrough · audit-ready · no vendor lock-in.
Federated insights, never federated data
Pulse Network shares calibrated benchmarks across customers ('companies your size estimate at 4w, ship at 7w') without data ever leaving your tenant.
Procurement-grade answers.
Most of these come up in evaluations. If yours isn’t here, ask directly.
Does Pulse store my source data?
Pulse keeps an indexed mirror with embedding vectors and structured entity extracts. Raw content is fetched on demand from the source system using your OAuth token. ACLs are revalidated per query.
What happens if I revoke a connector?
Within 60 seconds, all Pulse retrievals stop hitting that source. Within 24 hours, the indexed mirror is purged. Selective amnesia (#F63) supports finer-grained per-thread / per-doc removal.
Can I run Pulse against a single tool?
Yes. Connect Slack alone if that's where work lives. Each connector adds value independently; Pulse degrades gracefully when sources are missing.
How does BYOK pricing work?
On BYOK plans, Pulse charges a flat per-seat fee with no LLM markup. Your model bill goes directly to Anthropic / OpenAI / Azure. Suits enterprises with existing model contracts.
What MCP clients work?
Anything that speaks Model Context Protocol: Claude Desktop, Cursor, Cowork, ChatGPT (when their MCP support ships), and any custom client. We test deeply against Claude Desktop today.
Wire it in. See the difference in days.
Two weeks. Five connectors. A calibrated proof of concept against your actual data.