
agent-mail
MCP Agent Mail - Mail-like coordination layer for multi-agent workflows. Identities, inbox/outbox, file reservations, contact policies, threaded messaging, pre-commit guard, Human Overseer, static exports, disaster recovery. Git+SQLite backed. Python/FastMCP.
- Installs
- 78
- Last 24h
- —
- Trending rank
- —
- Security
- Not audited
Install
Works with every agent the skills CLI supports.
npx skills add https://github.com/dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations --skill agent-mail- 01
Install
Run the command in your project. The CLI asks which agents to install for.
- 02
Describe the task
Ask your agent for something this skill covers. It loads agent-mail automatically when relevant.
- 03
Review the result
The agent follows the SKILL.md steps, scripts and references bundled with the skill.
SKILL.md
The instructions your agent reads when the skill is triggered.
MCP Agent Mail
A mail-like coordination layer for coding agents exposed as an HTTP-only FastMCP server. Provides memorable identities, inbox/outbox, file reservation leases, contact policies, searchable message history, and Human Overseer messaging. Backed by Git (human-auditable artifacts) and SQLite (fast queries with FTS5).
Why This Exists
Without coordination, multiple agents:
- Overwrite each other's edits or panic on unexpected diffs
- Miss critical context from parallel workstreams
- Require humans to relay messages between tools
Agent Mail solves this with:
- Memorable identities (adjective+noun names like "GreenCastle")
- Advisory file reservations to signal editing intent
- Threaded messaging with importance levels and acknowledgments
- Pre-commit guard to enforce reservations at commit time
- Human Overseer for direct human-to-agent communication
Starting the Server
# Quickest way (alias added during install)
am
# Or manually
cd ~/projects/mcp_agent_mail
./scripts/run_server_with_token.sh
Default: http://127.0.0.1:8765
Web UI for humans: http://127.0.0.1:8765/mail
Core Concepts
Projects
Each working directory (absolute path) is a project. Agents in the same directory share a project namespace. Use the same project_key for agents that need to coordinate.
Agent Identity
Agents register with adjective+noun names (GreenCastle, BlueLake). Names are unique per project, memorable, and appear in inboxes, commit logs, and the web UI.
File Reservations (Leases)
Advisory locks on file paths or globs. Before editing files, reserve them to signal intent. Other agents see the reservation and can choose different work. The optional pre-commit guard blocks commits that conflict with others' exclusive reservations.
Contact Policies
Per-agent policies control who can message whom:
| Policy | Behavior |
|---|---|
open | Accept any message in the project |
auto (default) | Allow if shared context exists (same thread, overlapping reservations, recent contact) |
contacts_only | Require explicit contact approval first |
block_all | Reject all new contacts |
Messages
GitHub-Flavored Markdown with threading, importance levels (low, normal, high, urgent), and optional acknowledgment requirements. Images are auto-converted to WebP.
Essential Workflow
1. Start Session (One-Call Bootstrap)
macro_start_session(
human_key="/abs/path/to/project",
program="claude-code",
model="opus-4.5",
task_description="Implementing auth module"
)
Returns: {project, agent, file_reservations, inbox}
This single call: ensures project exists, registers your identity, optionally reserves files, fetches your inbox.
2. Reserve Files Before Editing
file_reservation_paths(
project_key="/abs/path/to/project",
agent_name="GreenCastle",
paths=["src/auth/**/*.ts", "src/middleware/auth.ts"],
ttl_seconds=3600,
exclusive=true,
reason="br-123"
)
Returns: {granted: [...], conflicts: [...]}
Conflicts are reported but reservations are still granted. Check conflicts and coordinate if needed.
3. Announce Your Work
send_message(
project_key="/abs/path/to/project",
sender_name="GreenCastle",
to=["BlueLake"],
subject="[br-123] Starting auth refactor",
body_md="Reserving src/auth/**. Will update session handling.",
thread_id="br-123",
importance="normal",
ack_required=true
)
4. Check Inbox Periodically
fetch_inbox(
project_key="/abs/path/to/project",
agent_name="GreenCastle",
limit=20,
urgent_only=false,
include_bodies=true
)
Or use resources for fast reads:
resource://inbox/GreenCastle?project=/abs/path&limit=20&include_bodies=true
5. Release Reservations When Done
release_file_reservations(
project_key="/abs/path/to/project",
agent_name="GreenCastle"
)
The Four Macros
Prefer macros for speed and smaller models. Use granular tools when you need fine control.
| Macro | Purpose |
|---|---|
macro_start_session | Bootstrap: ensure project → register agent → optional file reservations → fetch inbox |
macro_prepare_thread | Join existing conversation: register → summarize thread → fetch inbox context |
macro_file_reservation_cycle | Reserve files, do work, optionally auto-release when done |
macro_contact_handshake | Request contact permission, optionally auto-accept, send welcome message |
Beads Integration (br-### Workflow)
When using Beads for task management, keep identifiers aligned:
1. Pick ready work: br ready --json → choose br-123
2. Reserve files: file_reservation_paths(..., reason="br-123")
3. Announce start: send_message(..., thread_id="br-123", subject="[br-123] Starting...")
4. Work and update: Reply in thread with progress
5. Complete: br close br-123
release_file_reservations(...)
send_message(..., subject="[br-123] Completed")
Use br-### as:
- Mail
thread_id - Message subject prefix
[br-###] - File reservation
reason - Commit message reference
Beads Viewer (bv) Integration
Use bv's robot flags for intelligent task selection:
| Flag | Output | Use Case |
|---|---|---|
bv --robot-insights | PageRank, critical path, cycles | "What's most impactful?" |
bv --robot-plan | Parallel tracks, unblocks | "What can run in parallel?" |
bv --robot-priority | Recommendations with confidence | "What should I work on next?" |
bv --robot-diff --diff-since <ref> | Changes since commit/date | "What changed?" |
Rule of thumb: Use br for task operations, use bv for task intelligence.
Cross-Project Coordination
For frontend/backend or multi-repo projects:
Option A: Shared project_key
Both repos use the same project_key. Agents coordinate automatically.
Option B: Separate projects with contact links
# Backend agent requests contact with frontend agent
request_contact(
project_key="/abs/path/backend",
from_agent="GreenCastle",
to_agent="BlueLake",
to_project="/abs/path/frontend",
reason="API contract coordination"
)
# Frontend agent accepts
respond_contact(
project_key="/abs/path/frontend",
to_agent="BlueLake",
from_agent="GreenCastle",
accept=true
)
Pre-Commit Guard
Install the guard to block commits that conflict with others' exclusive reservations:
install_precommit_guard(
project_key="/abs/path/to/project",
code_repo_path="/abs/path/to/project"
)
Guard Features
- Composition-safe: Chain-runner preserves existing hooks in
hooks.d/ - Rename-aware: Checks both old and new paths for renames/moves
- NUL-safe: Handles paths with special characters
- Git-native matching: Uses Git wildmatch pathspec semantics
Set AGENT_NAME environment variable so the guard knows who you are.
Bypass in emergencies: AGENT_MAIL_BYPASS=1 git commit ...
Tools Reference
Project & Identity
| Tool | Purpose |
|---|---|
ensure_project(human_key) | Create/ensure project exists |
register_agent(project_key, program, model, name?, task_description?) | Register identity |
whois(project_key, agent_name) | Get agent profile with recent commits |
create_agent_identity(project_key, program, model) | Always create new unique agent |
Messaging
| Tool | Purpose |
|---|---|
send_message(project_key, sender, to, subject, body_md, ...) | Send message |
reply_message(project_key, message_id, sender, body_md) | Reply (preserves thread) |
fetch_inbox(project_key, agent, limit?, since_ts?, urgent_only?) | Get messages |
mark_message_read(project_key, agent, message_id) | Mark as read |
acknowledge_message(project_key, agent, message_id) | Acknowledge receipt |
search_messages(project_key, query) | FTS5 search |
summarize_thread(project_key, thread_id) | Extract key points and actions |
File Reservations
| Tool | Purpose |
|---|---|
file_reservation_paths(project_key, agent, paths, ttl?, exclusive?) | Reserve files |
release_file_reservations(project_key, agent, paths?) | Release reservations |
renew_file_reservations(project_key, agent, extend_seconds?) | Extend TTL |
force_release_file_reservation(project_key, agent, reservation_id) | Clear stale reservation |
Contact Management
| Tool | Purpose |
|---|---|
request_contact(project_key, from_agent, to_agent, reason?) | Request permission to message |
respond_contact(project_key, to_agent, from_agent, accept) | Accept/deny contact request |
list_contacts(project_key, agent_name) | List contact links |
set_contact_policy(project_key, agent_name, policy) | Set open/auto/contacts_only/block_all |
Resources (Fast Reads)
Use resources for quick, non-mutating reads:
resource://inbox/{agent}?project=<path>&limit=20&include_bodies=true
resource://thread/{thread_id}?project=<path>&include_bodies=true
resource://message/{id}?project=<path>
resource://file_reservations/{slug}?active_only=true
resource://project/{slug}
resource://projects
resource://agents/{project_key}
Search Syntax (FTS5)
"exact phrase"
prefix*
term1 AND term2
term1 OR term2
subject:login
body:"api key"
(auth OR login) AND NOT admin
Example: search_messages(project_key, '"auth module" AND error NOT legacy')
Web UI Features
Browse at http://127.0.0.1:8765/mail:
- Unified inbox across all projects
- Per-project search with FTS5
- Thread viewer with markdown rendering
- File reservations browser
- Human Overseer: Send high-priority messages to agents from the web UI
- Related Projects Discovery: AI-powered suggestions for linking repos
Human Overseer
Send direct messages to agents with automatic preamble:
- Messages marked as
highimportance - Bypasses contact policies
- Agents are instructed to pause current work, complete request, then resume
Static Mailbox Export
Export projects to portable, read-only bundles for auditors, stakeholders, or archives:
# Interactive wizard (recommended)
uv run python -m mcp_agent_mail.cli share wizard
# Manual export
uv run python -m mcp_agent_mail.cli share export --output ./bundle
# With signing
uv run python -m mcp_agent_mail.cli share export \
--output ./bundle \
--signing-key ./keys/signing.key
# Preview locally
uv run python -m mcp_agent_mail.cli share preview ./bundle
Export Features
- Ed25519 cryptographic signing
- Age encryption for confidential distribution
- Scrub presets:
standard(removes secrets) orstrict(redacts bodies) - Deploy to GitHub Pages or Cloudflare Pages via wizard
Disaster Recovery
# Save current state
uv run python -m mcp_agent_mail.cli archive save --label nightly
# List restore points
uv run python -m mcp_agent_mail.cli archive list --json
# Restore after disaster
uv run python -m mcp_agent_mail.cli archive restore <file>.zip --force
Mailbox Health (Doctor)
# Run diagnostics
uv run python -m mcp_agent_mail.cli doctor check
# Preview repairs
uv run python -m mcp_agent_mail.cli doctor repair --dry-run
# Apply repairs (creates backup first)
uv run python -m mcp_agent_mail.cli doctor repair
Checks: stale locks, database integrity, orphaned records, FTS sync, expired reservations.
Common Pitfalls
| Error | Fix |
|---|---|
| "sender_name not registered" | Call register_agent or macro_start_session first |
| "FILE_RESERVATION_CONFLICT" | Wait for expiry, coordinate, or use non-exclusive |
| "CONTACT_BLOCKED" | Use request_contact and wait for approval |
| Empty inbox | Check since_ts, urgent_only, verify agent name matches exactly |
Installation
# One-liner (recommended)
curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/mcp_agent_mail/main/scripts/install.sh?$(date +%s)" | bash -s -- --yes
# Custom port
curl -fsSL ... | bash -s -- --port 9000 --yes
# Change port after installation
uv run python -m mcp_agent_mail.cli config set-port 9000
Key Environment Variables
| Variable | Default | Description |
|---|---|---|
STORAGE_ROOT | ~/.mcp_agent_mail_git_mailbox_repo | Root for repos and SQLite DB |
HTTP_PORT | 8765 | Server port |
HTTP_BEARER_TOKEN | — | Static bearer token for auth |
LLM_ENABLED | true | Enable LLM for summaries/discovery |
CONTACT_ENFORCEMENT_ENABLED | true | Enforce contact policy |
Docker
docker build -t mcp-agent-mail .
docker run --rm -p 8765:8765 \
-e HTTP_HOST=0.0.0.0 \
-v agent_mail_data:/data \
mcp-agent-mail
Integration with Flywheel
| Tool | Integration |
|---|---|
| NTM | Agent panes coordinate via mail, dashboard shows inbox |
| BV | Task IDs become thread IDs, robot flags inform task selection |
| CASS | Search mail threads across sessions |
| CM | Extract procedural memory from mail archives |
| DCG | Mail notifies agents of blocked commands |
| RU | Coordinate multi-repo updates via cross-project mail |
FAQ
What does the agent-mail skill do?
MCP Agent Mail - Mail-like coordination layer for multi-agent workflows. Identities, inbox/outbox, file reservations, contact policies, threaded messaging, pre-commit guard, Human Overseer, static exports, disaster recovery. Git+SQLite backed. Python/FastMCP.
How do I install agent-mail?
Run npx skills add https://github.com/dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations --skill agent-mail in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/agent-mail/ (personal) or .claude/skills/agent-mail/ (project).
Which AI agents can use agent-mail?
Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.
Is agent-mail safe to install?
No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.
Who maintains agent-mail?
It is published by dicklesworthstone in dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations. It has 78 recorded installs.
More from dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations
ssh
dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations
SSH remote access - connections, tunnels, keys, file transfers. Use when connecting to servers, managing SSH keys, setting up port forwarding, or transferring files with scp/rsync.
ui-ux-polish
dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations
Open the skill to read its SKILL.md instructions.
gcloud
dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations
Google Cloud Platform CLI - manage GCP resources. Use when working with Compute Engine, Cloud Run, GKE, Cloud Functions, Storage, BigQuery, or other GCP services.
agent-swarm-workflow
dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations
Open the skill to read its SKILL.md instructions.
planning-workflow
dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations
Comprehensive markdown planning methodology for software projects. Use when starting a new project, creating implementation plans, or refining architecture before coding.
cass
dicklesworthstone/agent_flywheel_clawdbot_skills_and_integrations
Open the skill to read its SKILL.md instructions.
Popular ai & agents skills
google-agents-cli-publish
google/agents-cli
This skill should be used when the user wants to "publish an agent", "publish my ADK agent", "register an agent with Gemini", "publish to Gemini" (formerly Gemini Enterprise), or needs guidance on the agents-cli publish gemini command. Also use when the user wants to "manage agents in Agent Registry", "list/update/delete registered agents", or "register an MCP server". Covers ADK vs A2A registration modes, programmatic and interactive usage, flag reference, auto-detection from deployment metadata, Agent Registry fleet management, and troubleshooting. Part of the agents-cli skills suite. Do NOT use for deployment (use google-agents-cli-deploy).
google-agents-cli-eval
google/agents-cli
This skill should be used when the user wants to "run an evaluation", "evaluate my agent", "evaluate my ADK agent", "write an eval dataset", "analyze eval failures", "compare eval results", "optimize agent", or needs guidance on the Gemini platform eval methodology and the Quality Flywheel. Covers eval metrics, dataset schema, LLM-as-judge scoring, and common failure causes. Applies to any agents-cli project, whatever framework the agent is written in. Do NOT use for agent API code patterns (ADK: use google-agents-cli-adk-code), deployment (use google-agents-cli-deploy), or project scaffolding (use google-agents-cli-scaffold).
skill-creator
anthropics/skills
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
codex-pet
prime-skills/runcomfy-agent-skills
Codex Pet generator on RunComfy. Build a Codex-compatible Codex Pet spritesheet.webp + pet.json from a single reference image, drop it into ${CODEXHOME:-$HOME/.codex}/pets/ / and Codex picks it up as a custom Codex Pet next to the 8 built-ins. This skill produces the exact Codex Pet atlas Codex expects (1536x1872 PNG/WebP, 8 cols x 9 rows, 192x208 cells, 9 animation states — idle, running-right, running-left, waving, jumping, failed, waiting, running, review). Calls OpenAI GPT Image 2 edit ONCE via the local RunComfy CLI as runcomfy run openai/gpt-image-2/edit to produce a canonical Codex Pet pose, then assembles all 9 animation rows programmatically with ImageMagick micro-transforms — no Codex Pro, no $imagegen, no OPENAIAPIKEY required, only RUNCOMFYTOKEN. Triggers on "codex pet", "create codex pet", "make codex pet", "hatch codex pet", "/hatch image", "desktop pet codex", "codex pets", "spritesheet.webp", or any explicit ask to build a custom pet for OpenAI Codex.
full-output-enforcement
leonxlnx/taste-skill
Overrides default LLM truncation behavior. Enforces complete code generation, bans placeholder patterns, and handles token-limit splits cleanly. Apply to any task requiring exhaustive, unabridged output.
executing-plans
obra/superpowers
Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available