history-graph-protocol 0.3.0

An MCP server that gives AI agents a permanent, append-only causal history

History Graph Protocol (HGP)

An MCP server that gives AI agents a permanent, append-only causal history.

HGP records what an agent did and why — every operation, every causal link, and every piece of evidence behind each decision. Where memory systems (mem0, Zep, ChatGPT Memory) store what an agent knows, HGP stores what an agent did. Think of other memory systems as an agent's working memory; HGP is the agent's audit trail.

---

Core Concepts

Operations (Nodes)

Each operation is an immutable, typed, append-only record of an action taken by an agent. Four types:

Type	Meaning
artifact	A produced output (document, code, data)
hypothesis	A claim, decision, or inference
merge	Combining multiple prior operations
invalidation	Superseding or retracting a prior operation

Operations are never deleted or mutated. The graph only grows.

Causal Graph (DAG)

Operations are connected by directed edges:

• causal — A produced B (B depends on A)
• invalidates — A supersedes B (B is no longer current)

Example: an analysis produces a hypothesis, which is later revised by an invalidation.

graph LR
    A["op-1\nartifact\n(analysis)"]
    B["op-2\nhypothesis\n(decision)"]
    C["op-3\ninvalidation\n(revised decision)"]

    A -->|causal| B
    C -->|invalidates| B
    A -->|causal| C

Memory Tier

Each operation carries a memory tier that reflects access recency:

Tier	Meaning
short_term	Recently active; included in all queries
long_term	Older but still reachable; included in all queries
inactive	Not accessed recently; excluded from default queries, never deleted

Tiers are updated automatically based on access patterns and can be set manually via hgp_set_memory_tier. Valid tier values: short_term, long_term, inactive. See docs/tools-reference.md for details.

Evidence Trail (V3)

Operations can cite other operations as evidence, independently of the DAG edge structure. Evidence relations are stored separately and carry:

• relation — supports, refutes, context, method, or source
• scope — which aspect of the citing op this evidence applies to
• inference — free-text explanation of how the evidence was used

This allows full auditability: given any decision, you can reconstruct exactly what evidence it was based on and how each piece was interpreted.

---

Installation

Prerequisites: Python ≥ 3.12, uv

# With uv (recommended)
uv pip install history-graph-protocol

# With pip
pip install history-graph-protocol

Note: The PyPI package name is history-graph-protocol; the installed CLI command is hgp.

Run as MCP server (stdio transport):

hgp
# or
python -m hgp.server

Installing hooks

HGP ships hook scripts for Claude Code and Gemini CLI. Run once inside your project:

hgp install-hooks            # install both Claude Code and Gemini CLI hooks
hgp install-hooks --claude   # Claude Code only
hgp install-hooks --gemini   # Gemini CLI only

Hooks are installed into <repo_root>/.claude/hooks/ and <repo_root>/.gemini/hooks/. The command must be run from inside a git repository.

Hook enforcement policy

By default, hooks warn when native file tools are used instead of HGP equivalents. You can change this persistently with:

hgp hook-policy              # show current policy (default: advisory)
hgp hook-policy advisory     # warn only — native file tools allowed
hgp hook-policy block        # block native Write/Edit/write_file/replace

The policy is stored in <repo_root>/.hgp/hook-policy. The HGP_HOOK_BLOCK environment variable takes precedence over the file if set.

Note: If you installed hooks before this feature was added, run hgp install-hooks again to update them.

Storage

HGP stores its database and content-addressable blobs in <repo_root>/.hgp/ (gitignored). The server resolves the project root from the working directory at startup — it is bound to one repository per process.

Environment Variables

Variable	Default	Description
HGP_PROJECT_ROOT	(auto)	Override project root (default: nearest .git from cwd, or ~/.hgp/ if not in a repo)
HGP_GLOBAL_MODE	(unset)	Set to 1 to force legacy global store at ~/.hgp/ (all projects share one DB)
HGP_HOOK_BLOCK	0	Set to 1 to block native file tool calls (Write/Edit) instead of warning

MCP Client Configuration

Claude Code / Claude Desktop (claude_desktop_config.json or .claude/mcp.json):

{
  "mcpServers": {
    "hgp": {
      "command": "python",
      "args": ["-m", "hgp.server"]
    }
  }
}

The server is started from the project directory by the MCP host, so HGP_PROJECT_ROOT is typically not needed. Set it explicitly if the working directory at startup is not the project root.

---

Quick Start

These are MCP tool calls — not direct Python imports. Invoke them through your MCP client (Claude Code, Claude Desktop, or any MCP-compatible host).

# 1. Record an analysis operation (artifact)
hgp_create_operation(
    op_type="artifact",
    agent_id="agent-1",
    metadata={
        "description": "Analyzed error rate spike in service logs",
        "findings": "p99 latency exceeded 2s between 03:00-04:00 UTC",
    },
)
# → returns op_id: "op-abc123"

# 2. Record a decision (hypothesis) citing the analysis as evidence
hgp_create_operation(
    op_type="hypothesis",
    agent_id="agent-1",
    metadata={"description": "Root cause is database connection pool exhaustion"},
    parent_op_ids=["op-abc123"],
    evidence_refs=[
        {
            "op_id": "op-abc123",
            "relation": "supports",
            "scope": "latency correlation",
            "inference": "Latency spike aligns with connection pool saturation metrics",
        }
    ],
)
# → returns op_id: "op-def456"

# 3. Audit: what evidence did the decision cite?
hgp_get_evidence(op_id="op-def456")
# → lists all evidence records cited by op-def456

# 4. Audit: what decisions cited the original analysis?
hgp_get_citing_ops(op_id="op-abc123")
# → lists all ops that cited op-abc123 as evidence

---

Tool Index

Tool	Description
hgp_create_operation	Record a new operation; optionally attach payload, link parents, cite evidence
hgp_query_operations	Filter operations by type, agent, status, or memory tier
hgp_query_subgraph	Traverse ancestors or descendants from a root operation
hgp_acquire_lease	Acquire an optimistic lock on a subgraph before multi-step writes
hgp_validate_lease	Ping a lease to confirm it is still active (extends TTL by default)
hgp_release_lease	Release a lease explicitly after writing
hgp_set_memory_tier	Manually promote or demote an operation's memory tier
hgp_get_artifact	Retrieve binary payload from CAS by its object_hash
hgp_anchor_git	Link an operation to a Git commit SHA (requires full 40-char lowercase hex SHA)
hgp_reconcile	Run crash-recovery reconciler (use after unexpected shutdown)
hgp_get_evidence	List all operations a given op cited as evidence
hgp_get_citing_ops	Reverse lookup — list all ops that cited a given op as evidence
hgp_write_file	Write (create or overwrite) a file and record it as an artifact
hgp_append_file	Append content to a file and record as artifact
hgp_edit_file	Replace a unique string in a file and record as artifact
hgp_delete_file	Delete a file and record an invalidation operation
hgp_move_file	Move/rename a file; records invalidation of old path + new artifact
hgp_file_history	Return all HGP operations recorded for a given file path

→ Full API reference: docs/tools-reference.md → Usage patterns and examples: docs/usage-patterns.md

Token-sensitive sessions: mutation tools (hgp_create_operation, hgp_write_file, hgp_edit_file, hgp_append_file, hgp_delete_file, hgp_move_file) accept verbose=False to omit chain_hash and object_hash from responses, reducing per-call token overhead by ~73%. The default is verbose=True.