Agent Quickstart

Everything an AI agent or builder needs to integrate Anchord — MCP setup, working examples, and what to do when things need human review.

Anchord is read-only by design. It resolves identities and evaluates writes but never writes back to external systems. guard_write returns a decision; the caller executes any actual write.

Need to set up the console too? The Full Quickstart covers integrations, data exploration, duplicate management, and team setup.

1. Install the MCP server

Prerequisites

Node.js >= 20
An Anchord API key (get one at signup)

Run with npx (no install)

ANCHORD_API_KEY=your-key npx -y @anchord/mcp-server

The server starts over stdio and is ready for MCP clients. No clone or build step needed.

Cursor

Add to .cursor/mcp.json (workspace) or ~/.cursor/mcp.json (global):

{
  "mcpServers": {
    "anchord": {
      "command": "npx",
      "args": ["-y", "@anchord/mcp-server"],
      "env": {
        "ANCHORD_API_KEY": "<your-api-key>"
      }
    }
  }
}

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "anchord": {
      "command": "npx",
      "args": ["-y", "@anchord/mcp-server"],
      "env": {
        "ANCHORD_API_KEY": "<your-api-key>"
      }
    }
  }
}

2. Required environment variables

Variable	Required	Default	Description
ANCHORD_API_KEY	Yes	—	Bearer token from signup
ANCHORD_API_BASE_URL	No	https://api.anchord.ai	Production API. Use `http://localhost:8000` for local development only.

3. Smoke test

Verify your connection with two calls:

Ping

curl -s https://api.anchord.ai/api/ping \
  -H "Authorization: Bearer <your-key>"

Expected: {"status":"ok","tenant_id":"...","request_id":"..."}

Resolve one company

MCP tool call:

{
  "tool": "resolve_company",
  "arguments": { "domain": "acme.com" }
}

Or 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"}'

If you get {"status":"ok"} from ping and a response from resolve, you're connected.

4. Examples

A. Resolve a company by domain

Use when you have a company domain and need a stable AnchorID entity:

{
  "tool": "resolve_company",
  "arguments": { "domain": "stripe.com" }
}

Response (key fields):

{
  "status": "resolved",
  "entity_id": "ent_abc123",
  "confidence": 0.95,
  "safe_to_write": true,
  "reasons": ["DOMAIN_EXACT_MATCH"],
  "conflicts": [],
  "candidates": [],
  "attempt_id": "att_..."
  // match, targets, coverage omitted for brevity
}

Status	Meaning	What to do
resolved	High-confidence match found	Use `entity_id` as the canonical identity
not_found	No match in existing records	Safe to create a new record in the target system
needs_review	Ambiguous — multiple candidates	See section 5 below

B. Get an entity and link a source record

Fetch full entity details including all linked source records:

{
  "tool": "get_entity",
  "arguments": {
    "entity_id": "ent_abc123",
    "include": "source_records"
  }
}

Link a source record from another system to this entity:

{
  "tool": "link_source_record",
  "arguments": {
    "entity_id": "ent_abc123",
    "source_record_id": "sr_xyz789"
  }
}

C. Guard a write before executing it

Before your agent writes to an external system (CRM, database, etc.), ask Anchord to evaluate:

{
  "tool": "guard_write",
  "arguments": {
    "entity_id": "ent_abc123"
  }
}

Response:

{
  "allowed": true,
  "blocks": [],
  "entity_id": "ent_abc123",
  "confidence": 0.95,
  "active_links": 3,
  "audit_id": "aud_xyz789"
}

Result	Meaning	Agent action
allowed: true	Safe to write	Proceed with the write
allowed: false	Conflicts detected	Do not write; check `blocks` array for details

5. What to do when `needs_review`

resolve_* can return needs_review when Anchord finds multiple plausible matches and cannot auto-resolve with confidence. guard_write returns allowed: true/false with a blocks array instead.

Recommended agent pattern

Do not write. The data is ambiguous; writing could create duplicates or corrupt records.
Surface the candidates to the user. The response includes a candidates array with entity IDs and match scores.
Direct the user to the Review Queue. Link them to https://app.anchord.ai/app/queues/needs-review.
Retry later. Once a human resolves the ambiguity, subsequent resolve calls for the same record will return resolved.

"I tried to resolve 'Acme Corp' but Anchord found multiple possible matches (confidence too low to auto-resolve). A human needs to review this in the Review Queue. I'll retry after it's resolved."

6. Available MCP tools

Tool	What it does
resolve_company	Match a company to a canonical AnchorID entity
resolve_company_batch	Batch company resolution (max 200)
resolve_person	Match a person to a canonical AnchorID entity
resolve_person_batch	Batch person resolution (max 200)
get_entity	Fetch an AnchorID entity + linked source records
get_entity_export	Export the golden record for an AnchorID
link_source_record	Link a source record to an entity
unlink_source_record	Soft-delete a source record link
guard_write	Pre-write safety check (evaluation-only)
guard_write_batch	Batch pre-write safety check (max 200)
ingest_record	Ingest a source record into Anchord

7. Rate limits & quotas

API rate limit: 120 requests/min per tenant
Signup: 10/hour per IP, 3/day per email. Disposable email domains blocked.
Login: 20 attempts per 15 minutes per IP
Ingest: Up to 100 records per batch call
Resolve batch: Up to 200 items per batch call
Plan quotas enforce monthly/daily caps (see plan entitlements)

Staging environment (staging.api.anchord.ai): Available for testing. Data may be reset at any time. Use https://api.anchord.ai for production.

8. Recipes

Short, copy-paste flows for common agent patterns:

Resolve Before Write — ingest → resolve → guard → write (end-to-end)
Guard Before CRM Update — guard-check before touching the CRM
Needs-Review Escalation — detect ambiguity, queue for human review, retry