agenthold 0.1.3

Shared versioned state for multi-agent AI workflows

agenthold

Shared versioned state for multi-agent AI workflows. An MCP server that gives your agents a consistent, conflict-safe ground truth.

---

The problem

When two agents update the same value at the same time, the second write silently overwrites the first. No exception is raised. The value is wrong. The system keeps running.

In the example above, two agents both read a $10,000 budget and each allocate from it independently. The total committed reaches $15,000. The budget dict never complains.

This is not a race condition in the traditional sense. It is a read-modify-write conflict: each agent reads a value, does work, and writes back a result that assumes nothing changed in between. With multiple agents running concurrently, that assumption is always wrong.

---

How it works

agenthold solves this with optimistic concurrency control (OCC), the same mechanism Postgres uses in UPDATE ... WHERE version = N and DynamoDB uses in conditional writes.

Every value stored in agenthold has a version number. When an agent writes, it passes the version it read. If the stored version has changed since the read, the write is rejected with a ConflictError that includes the current value. The agent re-reads, recalculates, and retries.

The losing agent detects the conflict, re-reads the real remaining budget ($2,000), and adjusts its allocation. The total committed is always exactly $10,000. Every write is tracked.

OCC is the right fit for agent workflows because:

• Agents do work between reads and writes (network calls, LLM inference). You cannot hold a database lock across that work.
• Conflicts are rare. Retrying once is cheaper than acquiring a lock on every read.
• The retry logic is simple, explicit, and fully in the agent's control.

---

Quick start

Install

pip install agenthold

Add to your MCP client config

{
  "mcpServers": {
    "agenthold": {
      "command": "agenthold",
      "args": ["--db", "/path/to/state.db"]
    }
  }
}

Works with Claude Desktop, Cursor, Continue, and any MCP-compatible client.

Use as a Python library

from agenthold.store import StateStore
from agenthold.exceptions import ConflictError

store = StateStore("./state.db")

# Write a value (first write, no conflict check needed)
store.set("order-1234", "status", "received", updated_by="intake-agent")

# Read it back; always get the version number too
record = store.get("order-1234", "status")
print(record.value)    # "received"
print(record.version)  # 1

# Write with conflict detection; pass the version you read
try:
    store.set(
        "order-1234", "status", "processing",
        updated_by="fulfillment-agent",
        expected_version=record.version,  # rejected if another agent wrote first
    )
except ConflictError as e:
    # Another agent wrote between your read and write.
    # e.detail has the current version, value, and who wrote it.
    record = store.get("order-1234", "status")
    # ... recalculate and retry

---

Tools

agenthold exposes four tools over the Model Context Protocol.

agenthold_get

Read the current value of a state record.

{
  "namespace": "order-1234",
  "key": "status"
}

{
  "status": "ok",
  "namespace": "order-1234",
  "key": "status",
  "value": "processing",
  "version": 3,
  "updated_by": "fulfillment-agent",
  "updated_at": "2026-03-15T10:42:00.123456+00:00"
}

Returns {"status": "not_found"} if the key does not exist. No exception is raised.

---

agenthold_set

Write a value. Pass expected_version to enable conflict detection.

{
  "namespace": "order-1234",
  "key": "status",
  "value": "shipped",
  "updated_by": "logistics-agent",
  "expected_version": 3
}

Success:

{
  "status": "ok",
  "namespace": "order-1234",
  "key": "status",
  "version": 4,
  "previous_version": 3
}

Conflict (another agent wrote before you):

{
  "status": "conflict",
  "namespace": "order-1234",
  "key": "status",
  "expected_version": 3,
  "actual_version": 5,
  "actual_updated_by": "returns-agent",
  "actual_updated_at": "2026-03-15T10:42:01.456+00:00",
  "hint": "Call agenthold_get to read the current state, merge your changes, and retry with the new version."
}

expected_version patterns:

Value	Behaviour
Omitted	Unconditional write — overwrites any concurrent change without warning
0	Create-only guard — succeeds only if the key does not yet exist; conflicts if it does
N (from a prior agenthold_get)	Conflict-safe write — rejected if another agent wrote since your read

Pass expected_version=0 when initialising shared state that should only be written once. Omit it only for deliberate unconditional overwrites.

---

agenthold_list

List all current state records in a namespace.

{ "namespace": "order-1234" }

{
  "status": "ok",
  "namespace": "order-1234",
  "count": 3,
  "records": [
    { "key": "reserved",  "value": true,        "version": 2, "updated_by": "inventory-agent", "updated_at": "..." },
    { "key": "status",    "value": "processing", "version": 3, "updated_by": "fulfillment-agent", "updated_at": "..." },
    { "key": "total",     "value": 80.99,        "version": 2, "updated_by": "pricing-agent",   "updated_at": "..." }
  ]
}

---

agenthold_history

Read the version history of a state record, newest first. Useful for debugging coordination issues and auditing writes.

{
  "namespace": "order-1234",
  "key": "status",
  "limit": 5
}

{
  "status": "ok",
  "namespace": "order-1234",
  "key": "status",
  "history": [
    { "version": 3, "value": "processing",  "updated_by": "fulfillment-agent", "updated_at": "...", "event_type": "write" },
    { "version": 2, "value": "validated",   "updated_by": "validation-agent",  "updated_at": "...", "event_type": "write" },
    { "version": 1, "value": "received",    "updated_by": "intake-agent",      "updated_at": "...", "event_type": "write" }
  ]
}

Each entry includes an event_type field: "write" for normal writes, "delete" for deletion events. Delete tombstones have value: null. An empty history list means no writes have been recorded for this key — the key may not exist. Use agenthold_get to check current state.

---

Conflict detection

The read-modify-write pattern with expected_version is the core of agenthold. Here is the canonical retry loop:

from agenthold.store import StateStore
from agenthold.exceptions import ConflictError

store = StateStore("./state.db")

record = store.get("campaign", "budget")   # read once before doing work
do_expensive_work()                         # LLM call, API request, etc.

while True:
    new_value = compute_new_value(record.value)
    try:
        store.set(
            "campaign", "budget", new_value,
            updated_by="my-agent",
            expected_version=record.version,
        )
        break   # write succeeded
    except ConflictError:
        record = store.get("campaign", "budget")   # re-read and retry

Why this works: The version number is the contract. If the stored version has advanced since your read, another agent wrote first. You take the current value, recalculate, and try again. The number of retries is bounded by the number of concurrent writers. In practice, agents almost never conflict more than once.

Why not locks? Locks require a lease mechanism (what happens if the agent crashes holding a lock?), add latency on every read, and interact badly with the long I/O waits inherent in agent workflows. OCC pays a cost only when there actually is a conflict.

---

Examples

Two worked examples are included, each with a "before" and "after" script.

Order processing

Two agents update the same order record concurrently: an inventory agent marks it reserved and sets the status, a pricing agent applies a discount.

# The problem: one agent silently overwrites the other
uv run python examples/order_processing/without_agenthold.py

# The solution: conflict detection + retry
uv run python examples/order_processing/with_agenthold.py

Budget allocation

Two agents draw from a shared marketing budget. Without conflict detection, the budget is silently overcommitted. With agenthold, the losing agent re-reads the remaining balance and adjusts its allocation.

# The problem: $10,000 budget committed to $15,000 of spend
uv run python examples/budget_allocation/without_agenthold.py

# The solution: exact allocation, full audit trail
uv run python examples/budget_allocation/with_agenthold.py

---

Configuration

agenthold --db ./state.db

Flag	Default	Description
--db	./agenthold.db	Path to the SQLite database file. Use :memory: for an in-process store (testing only; data is lost when the process exits).

The database file is created automatically on first run. Back it up like any other SQLite file.

---

Development

git clone https://github.com/edobusy/agenthold.git
cd agenthold
uv sync --all-extras --dev

Run the tests:

uv run pytest tests/ -v

Check coverage:

uv run pytest tests/ --cov=agenthold --cov-report=term-missing

Lint and type-check:

uv run ruff check src/ tests/
uv run ruff format src/ tests/
uv run mypy src/

CI runs on Python 3.11 and 3.12 on every push to main.

---

Technical notes

These notes are here for engineers who want to understand the design decisions.

Why SQLite? SQLite is the right tool for this scope. It is zero-dependency, ships in the Python stdlib, and runs everywhere. WAL mode is enabled for forward-compatibility; the current implementation serialises all reads and writes through a single lock. Postgres adds an ops dependency with no benefit at this scale. The storage backend is behind a clean interface (StateStore) that can be swapped for Postgres when the need arises. Choosing a simple tool deliberately is not a limitation.

Why OCC instead of pessimistic locking? Locks require the holder to release them, which means the system must handle crashes, timeouts, and stale holders. That complexity is not worth it when conflicts are rare. OCC pays a cost only when a conflict actually occurs: one extra read and one retry. For multi-agent workflows where agents do significant work between reads and writes (LLM inference, API calls, tool execution), OCC is the correct choice.

What the versioning guarantees: Each key has a version that starts at 1 and increments by exactly 1 on every write. The state_history table is append-only and records every write before the live record is updated, so a crash between the two writes leaves history consistent. Deletions also write a tombstone entry to state_history (with event_type: "delete") before removing the live record, so the full lifecycle of a key is visible in history. The ordering guarantee is per-key, not global; two different keys can have their versions updated in any order.

What would change for production scale: Three things. First, replace SQLite with Postgres: better concurrent write throughput, replication, and managed hosting. The StateStore interface is already designed to make this a contained change. Second, add authentication: the current server trusts any caller on the stdio transport. A production deployment needs at minimum an API key check. Third, add the HTTP transport: the MCP SDK supports StreamableHTTPServer, which would let remote agents connect over the network instead of requiring a local process.

---

License

MIT. See LICENSE.