Skip to main content
Glama

Graph Unmerge

graph_unmerge
Destructive

Split a falsely merged entity into two, moving specified edges to the new entity. Use when entity resolution incorrectly merged distinct entities.

Instructions

Split a falsely merged entity back into two separate entities, redistributing specified edges. Use when entity resolution made a mistake (e.g. merged 'Anna' and 'Anne'). The original entity keeps every edge not listed in edges_to_move; the new entity gets the listed edges plus a fresh embedding stub (re-derive with graph_reembed). Logged to the audit trail with reason. Returns the IDs of both entities.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
entity_idYesThe merged entity ID to split
new_entity_nameYesName for the split-off entity
new_entity_typeYesType label for the split-off entity
edges_to_moveYesEdges to move to the new entity
reasonYesWhy splitting (logged in audit)

Implementation Reference

  • MCP tool registration for graph_unmerge with input schema (entity_id, new_entity_name, new_entity_type, edges_to_move, reason) and handler that delegates to client.unmerge() then logs to merge-audit.jsonl.
    // ─── Tool: graph_unmerge ───
    
    server.registerTool("graph_unmerge", {
      title: "Graph Unmerge",
      description:
        "Split a falsely merged entity back into two separate entities, redistributing specified edges. Use when entity resolution made a mistake (e.g. merged 'Anna' and 'Anne'). The original entity keeps every edge not listed in `edges_to_move`; the new entity gets the listed edges plus a fresh embedding stub (re-derive with graph_reembed). Logged to the audit trail with `reason`. Returns the IDs of both entities.",
      inputSchema: {
        entity_id: z.string().describe("The merged entity ID to split"),
        new_entity_name: z.string().describe("Name for the split-off entity"),
        new_entity_type: z.string().describe("Type label for the split-off entity"),
        edges_to_move: z.array(z.object({
          other_entity_id: z.string().describe("Entity on the other end of the edge"),
          relation_type: z.string().describe("Relationship type (e.g. WORKS_ON)"),
          direction: z.enum(["in", "out"]).describe("Direction relative to the entity being split"),
        })).describe("Edges to move to the new entity"),
        reason: z.string().describe("Why splitting (logged in audit)"),
      },
      annotations: { destructiveHint: true },
    }, async (args) => {
      try {
        const result = await client.unmerge(
          currentTenant(),
          args.entity_id,
          args.new_entity_name,
          args.new_entity_type as EntityType,
          args.edges_to_move.map((e) => ({
            ...e,
            relation_type: e.relation_type as RelationshipType,
          })),
          args.reason,
        );
    
        // Log to merge audit
        try {
          const auditDir = join(GRAPH_MEMORY_HOME, "logs");
          mkdirSync(auditDir, { recursive: true });
          const auditPath = join(auditDir, "merge-audit.jsonl");
          const entry = JSON.stringify({
            action: "unmerge",
            timestamp: new Date().toISOString(),
            ...result,
            reason: args.reason,
          });
          writeFileSync(auditPath, entry + "\n", { flag: "a" });
        } catch { /* audit logging is best-effort */ }
    
        return toolResult({ ...result, audit_logged: true });
      } catch (err) {
        return toolError(`graph_unmerge failed: ${err instanceof Error ? err.message : String(err)}`);
      }
    });
  • Core handler for unmerge: creates a new entity node in the same tenant, moves specified edges (outgoing/incoming) from the original to the new entity by copying properties and deleting the old edge, then returns counts of remaining and moved edges.
    async unmerge(
      tenantId: string,
      entityId: string,
      newEntityName: string,
      newEntityType: EntityType,
      edgesToMove: Array<{ other_entity_id: string; relation_type: RelationshipType; direction: "in" | "out" }>,
      reason: string,
    ): Promise<{
      original: { id: string; remaining_edges: number };
      new_entity: { id: string; name: string; moved_edges: number };
    }> {
      const newId = newEntityName.toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/^-|-$/g, "");
      const now = new Date().toISOString();
    
      // Create the new entity in the same tenant
      await this.run(
        `
        CREATE (n:Entity:\`${newEntityType}\` {
          tenant_id: $tenantId,
          id: $newId,
          name: $newEntityName,
          confidence: 0.5,
          times_mentioned: 1,
          first_seen: datetime($now),
          last_seen: datetime($now)
        })
        `,
        { tenantId, newId, newEntityName, now },
      );
    
      // Move specified edges (all participants must be in the same tenant)
      let movedCount = 0;
      for (const edge of edgesToMove) {
        if (edge.direction === "out") {
          const rows = await this.run(
            `
            MATCH (original:Entity {tenant_id: $tenantId, id: $entityId})-[r:\`${edge.relation_type}\`]->(other:Entity {tenant_id: $tenantId, id: $otherId})
            MATCH (newNode:Entity {tenant_id: $tenantId, id: $newId})
            WITH r, newNode, other, properties(r) AS props
            CREATE (newNode)-[newR:\`${edge.relation_type}\`]->(other)
            SET newR = props
            DELETE r
            RETURN count(newR) AS moved
            `,
            { tenantId, entityId, otherId: edge.other_entity_id, newId },
          );
          movedCount += Number(rows[0]?.["moved"] ?? 0);
        } else {
          const rows = await this.run(
            `
            MATCH (other:Entity {tenant_id: $tenantId, id: $otherId})-[r:\`${edge.relation_type}\`]->(original:Entity {tenant_id: $tenantId, id: $entityId})
            MATCH (newNode:Entity {tenant_id: $tenantId, id: $newId})
            WITH r, newNode, other, properties(r) AS props
            CREATE (other)-[newR:\`${edge.relation_type}\`]->(newNode)
            SET newR = props
            DELETE r
            RETURN count(newR) AS moved
            `,
            { tenantId, entityId, otherId: edge.other_entity_id, newId },
          );
          movedCount += Number(rows[0]?.["moved"] ?? 0);
        }
      }
    
      // Count remaining edges on original (tenant-scoped)
      const remainingRows = await this.run(
        `MATCH (n:Entity {tenant_id: $tenantId, id: $entityId})-[r]-() RETURN count(r) AS remaining`,
        { tenantId, entityId },
      );
      const remainingEdges = Number(remainingRows[0]?.["remaining"] ?? 0);
    
      return {
        original: { id: entityId, remaining_edges: remainingEdges },
        new_entity: { id: newId, name: newEntityName, moved_edges: movedCount },
      };
    }
  • Input schema for graph_unmerge defining entity_id, new_entity_name, new_entity_type, edges_to_move (array of {other_entity_id, relation_type, direction}), and reason.
    inputSchema: {
      entity_id: z.string().describe("The merged entity ID to split"),
      new_entity_name: z.string().describe("Name for the split-off entity"),
      new_entity_type: z.string().describe("Type label for the split-off entity"),
      edges_to_move: z.array(z.object({
        other_entity_id: z.string().describe("Entity on the other end of the edge"),
        relation_type: z.string().describe("Relationship type (e.g. WORKS_ON)"),
        direction: z.enum(["in", "out"]).describe("Direction relative to the entity being split"),
      })).describe("Edges to move to the new entity"),
      reason: z.string().describe("Why splitting (logged in audit)"),
    },
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

The description adds significant behavioral detail beyond the destructiveHint annotation: it explains that edges are redistributed, the original entity keeps unmoved edges, the new entity gets a stub embedding, and the action is logged to an audit trail with a reason. No contradictions with annotations.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is two sentences, front-loaded with the main action and purpose. Every sentence adds value: the first defines the operation and usage context, the second provides critical details about edge redistribution, embedding, audit logging, and return value.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the complexity of a destructive graph operation with 5 parameters and no output schema, the description is remarkably complete. It explains the behavior, side effects, return value (IDs of both entities), and follow-up action (re-embedding). No critical gaps.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

All 5 parameters have schema descriptions (100% coverage), so baseline is 3. The description adds value by explaining that edges_to_move specifies which edges to move and that the original entity keeps the rest. It also notes that reason is logged to audit, which is not clear from schema. However, new_entity_name and new_entity_type parameters are not elaborated beyond schema.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the verb 'split' and the resource 'falsely merged entity'. It distinguishes from sibling tools like graph_merge and graph_delete by specifying it's for undoing a mistaken merge. The example with 'Anna' and 'Anne' reinforces the purpose.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicitly says 'Use when entity resolution made a mistake', providing clear guidance on when to use. It also advises re-deriving embeddings with graph_reembed after splitting, which is a helpful follow-up step. Implicitly, it should not be used for correct merges.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Install Server

Other Tools

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/stevepridemore/graph-memory'

If you have feedback or need assistance with the MCP directory API, please join our Discord server