relationship-manager
Ingests Fathom recorded calls as a source, adding call history to person timelines and relationship context.
Ingests Gmail email into the relationship graph, enabling identity resolution, timeline merging, and prep briefs based on email history.
Syncs Slack data via local connectors into the relationship graph as a source for identity resolution and activity context.
Syncs Stripe data via local connectors into the relationship graph as a source for identity resolution and activity context.
Ingests WhatsApp messages and identifiers (JID) to resolve contacts and include WhatsApp conversations in person timelines and briefs.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
relationship-manager
One graph of the people you actually talk to — and an agent API on top of it.
Your relationships are spread across Gmail, WhatsApp, your calendar and recorded calls. Every tool that tries to "manage" them wants you to type data in — the one thing you will never do.
relationship-manager reads the graph you already produced: it resolves the same human across
email, phone, WhatsApp JID and meeting attendee, keeps what it learns as evidence-backed
facts, and exposes the whole thing to agents first — one remote MCP endpoint, a REST API
and a CLI generated from a single tool contract.
Humans get a calm surface for the two jobs that actually need a human: deciding what's true and deciding who to reach out to.
What it does
Resolves identity across sources.
person_identifiersis the authoritative index — email, phone, WhatsApp JID, meeting-attendee name. One human, one row, no duplicates invented.Keeps evidence, not vibes. Every fact carries the observations that produced it and a band scored in code:
VERIFIED(may be written),PROBABLE/POSSIBLE(offered for your call), below that (not stored at all). The model never supplies a confidence number.Prepares you.
prep_briefis deterministic and free — who they are, how you're reachable, what's on file, recent email, meetings, WhatsApp, calls.Knows who to reconnect with. Cohort-ranked, suppression-respecting, with the history it was derived from.
Never sends anything. Agents can propose outreach; a human approves it. Writes are a different class of thing.
Related MCP server: openoutbound
Architecture
flowchart LR
subgraph Sources["Sources"]
G[Gmail]
W[WhatsApp]
C[Calendar]
F[Fathom calls]
S[Slack / Stripe]
end
subgraph Local["Local ingest (optional)"]
L[connectors → POST /api/sync]
end
subgraph Edge["Cloudflare Worker"]
A[apps/api: REST + MCP]
D[(D1: the relationship graph)]
U[apps/web: UI]
end
subgraph Agents["Agents"]
M1[Claude Code]
M2[Codex]
M3[Anything MCP]
end
G --> L --> A
W --> L
C --> L
F --> L
S --> L
A --- D
U --- A
M1 -->|MCP| A
M2 -->|MCP| A
M3 -->|MCP| AOne definition drives all three surfaces: packages/core/src/tools.ts. The MCP server, the REST
router and the CLI register from that list, so the tool you read about is the tool an agent calls.
Monorepo layout
apps/api Cloudflare Worker — REST + remote MCP + auth + D1
apps/web Vite + React UI (AIOS design tokens, light/dark)
packages/core evidence ledger, identity resolution, the tool contract (+ tests)
packages/seed importers: real corpora + live SQLite → migrations/seed.sql
packages/cli `rel` — talk to the deployed graph from a terminal; stdio MCP bridge
skill/ the agent skill bundle (teaches an agent how to use the tools well)
connectors/ (phase 2) local Python ingest → POST /api/syncTools (the agent surface)
Tool | Scope | What it does |
| read | Find people by name, email, company or domain |
| read | Full record: identity, evidence-backed facts, identifiers, counts |
| read | Deterministic markdown brief before you talk to someone |
| read | Merged chronology: email, meetings, WhatsApp, calls |
| read | Facts that are settled and facts awaiting a human decision |
| read | Ranked reconnect list with cohort, signal and draft |
| read | Honest per-source freshness — connected, stale, error |
| write | The only path for agent-derived facts; evidence bands decide what happens |
| write | Accept or dismiss a proposed fact (accept freezes the field as human-held) |
| write | Record an outreach attempt and a follow-up date |
| write | Queue a draft for human approval — agents cannot send |
Quick start
npm install
cp .dev.vars.example .dev.vars # LOGIN_PASSWORD, SESSION_SECRET
# build + seed + deploy
npm run build
npm run seed:build # reads reference/ → packages/seed/out/seed.sql
npx wrangler d1 create relationship-manager # paste database_id into wrangler.jsonc
npm run db:migrate && npm run seed:push
npx wrangler secret put LOGIN_PASSWORD
npx wrangler secret put SESSION_SECRET
npm run deployLocal development (UI against the real API):
npm run dev # Vite dev server, proxies /api → wrangler dev
npx wrangler dev # the API on :8787Connect an agent
# Claude Code
claude mcp add --transport http relationship-manager https://<your-worker>/mcp \
--header "Authorization: Bearer rel_..."
# Codex (~/.codex/config.toml)
[mcp_servers.relationship-manager]
url = "https://<your-worker>/mcp"
http_headers = { Authorization = "Bearer rel_..." }Then ask: "who should I reconnect with this week, and why?" — the agent calls
reconnect_queue and prep_brief, and every call shows up in the Agents screen.
Safety
Reads by default. Writes are opt-in per agent key (
scopes: read|write), validated server-side in one place, and logged.Facts need evidence.
score_evidence()decides the band; belowPOSSIBLEnothing is stored.No sending. Agents can draft and propose; a human approves. Outreach is logged, never fired.
Your data never enters this repo. The graph lives in your own Cloudflare D1 database and in local files (
reference/, the generatedpackages/seed/out/) — all gitignored. The repository is the engine, the surfaces and the deploy path, and nothing else. Run it in your own account, behind a password.
License
MIT — see LICENSE.
Part of the everyai-com agent stack
distillory — local-first memory engine that reasons at ingestion
agent-ready — turn any backend into an MCP server, API and CLI
agentprofile — one agent identity across every tool
primer — live business context injected into any agent
plainsync — local-first Markdown workspace for humans + agents
argus — cloud-native software verification
Built by Phanindra Reddy · magicteams.ai
License
Apache-2.0 — see LICENSE.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Personal CRM operated by AI agents. Read relationship context before acting, write back after.
Find LinkedIn prospects, draft outreach, and run human-paced campaigns from your AI agent.
Company brain for AI agents — temporal knowledge graph search, exploration, and durable memory.
Your company's brain for AI agents. Cited, permission-aware knowledge across every system.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceEnables agents to query and reason over a unified graph of a person's real relationships, resolved across email, WhatsApp, calendar, and calls, with evidence-backed facts and deterministic briefs.MIT
- AlicenseNot gradedqualityBmaintenanceEnables agents to find leads, read buying signals, write and send email and LinkedIn outreach, handle replies, and report results, with built-in safety gates, approvals, limits, and compliance controls.13Apache 2.0
- AlicenseNot gradedqualityCmaintenanceEnables AI agents to operate a personal CRM by reading relationship context and writing back. Provides 80 tools for daily briefings, semantic search, person digests, suggested actions, interaction logging, and full CRUD across contacts, notes, tasks, reminders, calls, gifts, journal entries, addresses, relationships, places, debts, documents, and life events.2MIT

Heliyos MCP serverofficial
AlicenseAqualityBmaintenanceEnables AI agents to work in a CRM workspace by finding and updating records, reading timelines and communications, managing notes, tasks, reviews, workers, scouts, and knowledge base content, and delegating jobs while keeping important actions subject to human review.73297 npmMIT