For agent builders

The identity layer your agent calls before every write

Anchord is a read-only identity + trust API. Your agent calls it before writing to HubSpot, Salesforce, Stripe, or any other system. Anchord resolves the record to a canonical identity, evaluates the write, and returns a decision. It never writes back.

Get Started Free Agent Quickstart

Try it now

Sign up (free, no credit card), save your API key, and run your first resolve:

Via MCP (Cursor / Claude Desktop)

// MCP tool call — works in Cursor, Claude Desktop, any MCP client
{
  "tool": "resolve_company",
  "arguments": { "domain": "acme.com" }
}

// Response
{
  "status": "resolved",
  "entity_id": "ent_abc123",
  "confidence": 0.95,
  "reasons": ["DOMAIN_EXACT_MATCH"]
}

Via curl

curl -s -X POST https://api.anchord.ai/api/v1/resolve/company \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"domain": "acme.com", "name": "Acme Corp"}'

MCP setup: Agent Quickstart

Core workflow: ingest → resolve → guard

Three API calls between your agent and the external system it's about to write to.

1

Ingest

Push records from any source via POST /api/v1/ingest/batch. Or connect Salesforce, HubSpot, and Stripe through built-in OAuth — records flow in automatically via sync and webhooks.

2

Resolve

Call POST /api/v1/resolve/company or /resolve/person. Anchord matches the record to an AnchorID — the canonical identity record for a Company or Person — using rules-based scoring and returns one of:

resolved

High-confidence match. Use the returned entity ID.

not_found

No match found. Safe to create a new record in the target system.

needs_review

Ambiguous. Do not write — escalate to a human.

3

Guard

Before writing, call POST /api/v1/guard/write with the entity ID. Anchord evaluates the write and returns:

allowed: true

Safe to proceed. Your agent executes the write.

allowed: false

Check the blocks array for specific codes (e.g. ENTITY_NOT_FOUND, NO_CANONICAL_LINK). Do not write.

Anchord never writes to external systems. The guard endpoint is evaluation-only — the caller decides.

Decision contract

Anchord answers identity and safety questions. Your agent decides what to do with the answers.

Endpoint Returns Agent action
resolve/company
resolve/person		 resolved + entity_id Proceed to guard, then write to the resolved entity
not_found Proceed to guard; safe to create a new record
needs_review Stop. Do not write. Log the attempt_id and surface to human
guard/write allowed: true Safe to write. Agent executes the write to the external system
allowed: false + blocks Do not write. Check the blocks array for specific codes

Resolve owns identity state. Guard evaluates whether a proposed write should proceed. The caller always performs the actual write.

What happens on needs_review

When Anchord finds multiple plausible matches and can't auto-resolve, it returns needs_review instead of guessing. This is the key safety mechanism.

  1. 1

    Agent stops

    The agent does not write. It logs the attempt_id and surfaces the ambiguity to the user.

  2. 2

    Human reviews

    The Anchord console shows the incoming record, candidate matches with confidence scores, and a recommendation. One click to confirm, dismiss, defer, or create a new entity.

  3. 3

    Agent retries

    After resolution, the same resolve call returns resolved with a stable AnchorID. The agent can proceed.

When to use Anchord

Any time your agent is about to write to an external system and identity matters — which is most of the time. See all scenarios →

Before updating a HubSpot company or contact

Resolve by domain or email first. Block the write if there's a conflict or near-duplicate.

Before creating a Salesforce account or contact

Catch duplicates Salesforce's own rules miss. Escalate ambiguous matches to a human instead of creating a new record.

Before syncing Stripe customer data into a CRM

When webhook events fire, resolve the customer identity before your pipeline pushes it downstream.

Preventing duplicate records across systems

Custom sources work too. Register your system, ingest records, resolve, and guard — same workflow regardless of the target.

What you get

REST API + 11-tool MCP server

JSON API for any HTTP client. MCP server with 11 tools for Cursor, Claude Desktop, and any MCP-compatible agent.

Built-in connectors

Salesforce, HubSpot, and Stripe connect via OAuth. Nightly syncs + webhooks pull data automatically.

Duplicate detection + merge

Deterministic clustering surfaces near-duplicate entities with confidence scores. Merge with dry-run preview.

Console for humans

Review Queue, Data Explorer, Identity Quality, Integration Health, and Matching Analytics reports. Agents use the API; humans monitor, tune, and resolve edge cases.

Start building

Sign up, get your API key, and run your first resolve call. Free plan, no credit card.

Create your account Agent Quickstart
Full Quickstart · Recipes · Integrations · API Reference