API Reference - AgentFront

CodeCall exposes four meta-tools that replace direct tool access. This page documents the complete API for each tool with request/response schemas and examples.

codecall:search

Search for tools by natural language query using semantic embeddings.

Request Schema

{
  "tool": "codecall:search",
  "input": {
    "queries": ["string (required) - Natural language search queries"],
    "topK": "number (optional, default: 10) - Maximum results per query",
    "appIds": ["string (optional) - Limit to specific apps"],
    "excludeToolNames": ["string (optional) - Tool names to exclude"],
    "minRelevanceScore": "number (optional) - Minimum relevance score threshold"
  }
}

Response Schema

{
  "tools": [
    {
      "name": "string - Tool name (e.g., 'users:list')",
      "appId": "string - Owning app ID",
      "description": "string - Tool description",
      "relevanceScore": "number - Relevance score (0-1)",
      "matchedQueries": ["string - Queries that matched this tool"]
    }
  ],
  "totalAvailableTools": "number - Total tools in index",
  "warnings": [
    {
      "type": "excluded_tool_not_found | no_results | low_relevance",
      "message": "string - Warning message",
      "affectedTools": ["string (optional) - Affected tool names"]
    }
  ]
}

Examples

{
  "tool": "codecall:search",
  "input": {
    "queries": ["get user profile information"],
    "topK": 5
  }
}

{
  "tool": "codecall:search",
  "input": {
    "queries": ["list all invoices"],
    "topK": 10,
    "appIds": ["billing", "payments"]
  }
}

Response Example

{
  "tools": [
    {
      "name": "users:getProfile",
      "appId": "user-service",
      "description": "Get user profile by ID with full details",
      "relevanceScore": 0.94,
      "matchedQueries": ["get user profile information"]
    },
    {
      "name": "users:list",
      "appId": "user-service",
      "description": "List all users with pagination and filtering",
      "relevanceScore": 0.78,
      "matchedQueries": ["get user profile information"]
    },
    {
      "name": "auth:whoami",
      "appId": "auth-service",
      "description": "Get current authenticated user info",
      "relevanceScore": 0.72,
      "matchedQueries": ["get user profile information"]
    }
  ],
  "totalAvailableTools": 156
}

Search Strategy

CodeCall supports two embedding strategies:

Strategy	Description	Best For
`tfidf`	TF-IDF text matching (default)	Fast, no model loading, works offline
`ml`	Semantic embeddings via VectoriaDB	Better relevance, requires model

CodeCallPlugin.init({
  embedding: {
    strategy: 'tfidf',  // or 'ml'
  },
});

codecall:describe

Get detailed schemas and documentation for specific tools.

Request Schema

{
  "tool": "codecall:describe",
  "input": {
    "toolNames": ["string (required) - Array of tool names to describe"]
  }
}

Response Schema

{
  "tools": [
    {
      "name": "string - Tool name",
      "appId": "string - Owning app ID",
      "description": "string - Full description",
      "inputSchema": { /* JSON Schema */ },
      "outputSchema": { /* JSON Schema or null */ },
      "annotations": {
        "readOnlyHint": "boolean (optional)",
        "destructiveHint": "boolean (optional)",
        "idempotentHint": "boolean (optional)",
        "openWorldHint": "boolean (optional)"
      },
      "usageExamples": [
        {
          "description": "string - What this example demonstrates",
          "code": "string - JavaScript snippet using callTool()"
        }
      ]
    }
  ],
  "notFound": ["string (optional) - Tool names that weren't found, or that the access policy withholds"]
}

usageExamples returns up to 5 entries per tool. Each entry has a description and a code snippet that calls the tool via callTool(...) so the LLM can copy the pattern directly into a codecall:execute script. They are the tool’s own examples when it declares any, and nothing else. A tool without examples gets one generated example whose arguments are built only from its input schema’s properties (the required properties, the query-like or first filter property for a search, the real pagination properties for a list, the first value of an enum), or {} when none fits; an argument the schema does not declare is never shown. Values stay within the property’s enum, const, minimum/maximum and minLength/maxLength; a pattern or format is not generated, so an optional query that has one is left out and a required property that has one gets a plain placeholder string. A list tool without pagination properties gets only its required properties, and an update adds the first two non-id properties.

Example

{
  "tool": "codecall:describe",
  "input": {
    "toolNames": ["users:list", "users:getById"]
  }
}

{
  "tools": [{
    "name": "users:list",
    "appId": "user-service",
    "description": "List users with optional filtering and pagination",
    "inputSchema": {
      "type": "object",
      "properties": {
        "status": {
          "type": "string",
          "enum": ["active", "inactive", "pending"],
          "description": "Filter by user status"
        },
        "role": {
        "type": "string",
        "enum": ["admin", "user", "guest"],
        "description": "Filter by role"
        },
        "limit": {
          "type": "number",
          "minimum": 1,
          "maximum": 100,
          "default": 20,
          "description": "Maximum results per page"
        },
        "offset": {
          "type": "number",
          "minimum": 0,
          "default": 0,
          "description": "Pagination offset"
        }
      }
    },
    "outputSchema": {
      "type": "object",
      "properties": {
        "users": {
          "type": "array",
          "items": {"$ref": "#/definitions/User"}
        },
        "total": {"type": "number"},
        "hasMore": {"type": "boolean"}
      }
    },
    "annotations": {
      "readOnlyHint": true
    },
    "usageExamples": [
      {
        "description": "List active users with pagination",
        "code": "const result = await callTool(\"users:list\", { status: \"active\", limit: 50 });"
      },
      {
        "description": "List admin users",
        "code": "const result = await callTool(\"users:list\", { role: \"admin\" });"
      },
      {
        "description": "List all users",
        "code": "const result = await callTool(\"users:list\", {});"
      }
    ]
  }],
  "notFound": []
}

Security Note

codecall:describe will not return information about CodeCall meta-tools themselves. Attempting to describe codecall:execute returns the tool in notFound.

codecall:execute

Execute a JavaScript (AgentScript) plan that orchestrates multiple tools.

Request Schema

{
  "tool": "codecall:execute",
  "input": {
    "script": "string (required) - JavaScript code to execute",
    "allowedTools": ["string (optional) - Whitelist of allowed tool names"]
  }
}

Response Schema

Success:

{
  "status": "ok",
  "result": "any - Return value from script",
  "logs": ["string (optional) - Console output if enabled"]
}

Error Responses:

Syntax Error

{
  "status": "syntax_error",
  "error": {
    "message": "Failed to parse AgentScript code: Unexpected token (5:12)",
    "location": {"line": 5, "column": 12}
  }
}

Validation Error (AST)

{
  "status": "illegal_access",
  "error": {
    "kind": "IllegalBuiltinAccess",
    "message": "Identifier 'eval' is not allowed in AgentScript"
  }
}

The lines an issue names (FORBIDDEN_LOOP (line 3): …) are the script’s own lines. A line that belongs to no line of the script is left out.

Runtime Error

{
  "status": "runtime_error",
  "error": {
    "source": "script",
    "message": "Cannot read property 'name' of undefined",
    "name": "TypeError"
  }
}

Tool Error

{
  "status": "tool_error",
  "error": {
    "source": "tool",
    "toolName": "users:list",
    "message": "Tool \"users:list\" execution failed",
    "code": "EXECUTION"
  }
}

A failing tool call the script didn’t catch (or rethrew), through callTool() or a namespace method, including a call the access policy refused (ACCESS_DENIED). An error of the script’s own, even after it caught a tool failure, is runtime_error. The result doesn’t carry the call’s arguments: the schema keeps toolInput, deprecated and never set, since the sandbox resolves sidecar references in them before the tool sees them.

Timeout

{
  "status": "timeout",
  "error": {
    "message": "Script execution timed out after 3500ms"
  }
}

Example

{
  "tool": "codecall:execute",
  "input": {
    "script": "const users = await callTool('users:list', { status: 'active' });\nconst admins = users.users.filter(u => u.role === 'admin');\nreturn { adminCount: admins.length, admins: admins.map(a => a.email) };"
  }
}

{
  "status": "ok",
  "result": {
    "adminCount": 3,
    "admins": ["alice@example.com", "bob@example.com", "charlie@example.com"]
  }
}

AgentScript Guide

For the complete scripting API (callTool, getTool, codecallContext, __safe_parallel), error handling patterns, and best practices, see the dedicated AgentScript guide.

codecall:invoke

Direct tool invocation without JavaScript execution. Useful for simple single-tool calls.

Request Schema

{
  "tool": "codecall:invoke",
  "input": {
    "tool": "string (required) - Tool name to invoke",
    "input": { /* (required) - Input to pass to the tool */ }
  }
}

Response Schema

codecall:invoke returns the standard MCP CallToolResult — the same envelope a direct tools/call produces. Errors are surfaced via isError: true plus a textual content entry; there is no separate { status, error } wrapper. Success:

{
  "content": [
    { "type": "text", "text": "..." }
  ],
  "structuredContent": "any (optional) - structured tool output when the tool declares an outputSchema",
  "isError": false
}

Error:

{
  "content": [
    { "type": "text", "text": "Tool \"users:delete\" not found. Use codecall:search to discover available tools." }
  ],
  "isError": true
}

Example

{
  "tool": "codecall:invoke",
  "input": {
    "tool": "users:getById",
    "input": { "id": "user-123" }
  }
}

{
  "content": [
    {
      "type": "text",
      "text": "{\"id\":\"user-123\",\"email\":\"alice@example.com\",\"name\":\"Alice Smith\",\"role\":\"admin\",\"status\":\"active\"}"
    }
  ],
  "structuredContent": {
    "id": "user-123",
    "email": "alice@example.com",
    "name": "Alice Smith",
    "role": "admin",
    "status": "active"
  },
  "isError": false
}

When to Use Invoke vs Execute

Use `codecall:invoke`	Use `codecall:execute`
Single tool call	Multiple tool calls
No data transformation needed	Filter/join/transform results
Latency-sensitive (no VM overhead)	Complex orchestration logic
Simple CRUD operations	Conditional workflows

Error Codes Reference

Script Execution Errors

Code	Status	Description	Recovery
`SYNTAX_ERROR`	`syntax_error`	JavaScript syntax error	Fix syntax at indicated location
`VALIDATION_ERROR`	`illegal_access`	AST validation failed	Remove blocked construct
`SELF_REFERENCE_BLOCKED`	`illegal_access`	Tried to call codecall:* tool	Use regular tools only
`TOOL_NOT_FOUND`	`tool_error`	Tool doesn’t exist	Check tool name spelling
`ACCESS_DENIED`	`tool_error`	Tool withheld by the access policy, or not in the script’s `allowedTools`	Use a tool `codecall:search` returns
`EXECUTION_ERROR`	`runtime_error`	Runtime error in script	Debug script logic
`TIMEOUT`	`timeout`	Exceeded time limit	Optimize or increase timeout
`TOOL_EXECUTION_ERROR`	`tool_error`	Tool threw an error	Check tool input

Error Handling

For error handling patterns including retry, fallback, and partial success strategies, see the AgentScript Guide.

Debugging Guide

Using console.log

// Enable in config
CodeCallPlugin.init({
  vm: { allowConsole: true }
});

// In script
const users = await callTool('users:list', { limit: 10 });
console.log('Fetched users:', users.length);

for (const user of users) {
  console.log('Processing:', user.id, user.email);
}

// Logs appear in response.logs array

Interpreting Error Messages

Maximum iteration limit exceeded

Cause: Loop ran more than maxSteps times Fix: Use pagination or filter data before looping

// Bad: looping over potentially large dataset
for (const item of items) { ... }

// Good: paginate or limit
const page = items.slice(0, 100);
for (const item of page) { ... }

Maximum tool call limit exceeded

Cause: Script called more than maxToolCalls tools Fix: Batch operations or use __safe_parallel

// Bad: one tool call per item
for (const id of ids) {
  await callTool('users:get', { id });
}

// Good: batch fetch
const users = await callTool('users:getBatch', { ids });

Script execution timed out

Cause: Script ran longer than timeoutMs Fix: Optimize, use less data, or increase timeout

// Check preset timeout
// locked_down: 2s, secure: 3.5s, balanced: 5s
CodeCallPlugin.init({
  vm: { preset: 'balanced', timeoutMs: 8000 }
});

Identifier 'X' is not allowed

Cause: Used a blocked identifier (eval, require, etc.) Fix: Use allowed alternatives

// Blocked
eval('code');
require('module');
setTimeout(fn, 100);

// Allowed
// Use callTool for external operations
await callTool('code:run', { script: 'code' });

Development Mode

For easier debugging during development:

CodeCallPlugin.init({
  vm: {
    preset: 'experimental',  // Longer timeouts, more iterations
    allowConsole: true,
  },
});

Testing Strategy

Unit Testing AgentScript

Use the Enclave library directly to test script execution in isolation:

import { Enclave } from '@enclave-vm/core';

describe('AgentScript', () => {
  let enclave: Enclave;

  beforeEach(() => {
    enclave = new Enclave();
  });

  afterEach(() => {
    enclave.dispose();
  });

  it('should execute basic script', async () => {
    const result = await enclave.run(`
      return 1 + 1;
    `);

    expect(result.success).toBe(true);
    expect(result.value).toBe(2);
  });

  it('should call tools', async () => {
    const mockToolHandler = jest.fn().mockResolvedValue({ users: [] });
    const enclaveWithTools = new Enclave({ toolHandler: mockToolHandler });

    const result = await enclaveWithTools.run(`
      const data = await callTool('users:list', {});
      return data.users.length;
    `);

    expect(mockToolHandler).toHaveBeenCalledWith('users:list', {});
    expect(result.value).toBe(0);

    enclaveWithTools.dispose();
  });
});

Integration Testing

For end-to-end tests, set up a FrontMCP app with CodeCallPlugin and invoke the meta-tools through your MCP test client:

import { App, Tool, ToolContext } from '@frontmcp/sdk';
import { CodeCallPlugin } from '@frontmcp/plugin-codecall';

@App({ id: 'test-app', plugins: [CodeCallPlugin.init()] })
class TestApp {
  @Tool({ name: 'users:list' })
  async listUsers(ctx: ToolContext) {
    return [
      { id: '1', name: 'Alice' },
      { id: '2', name: 'Bob' },
    ];
  }
}

// Use your MCP test client to call codecall:execute
// and verify the response status and result

Testing Security Rules

describe('Security', () => {
  it('should block eval', async () => {
    const enclave = new Enclave();
    const result = await enclave.run(`
      eval('1 + 1');
    `);

    expect(result.success).toBe(false);
    expect(result.error?.message).toContain('eval');

    enclave.dispose();
  });

  it('should enforce iteration limit', async () => {
    // Note: maxIterations is the Enclave-direct option (max loop iterations
    // per loop). The CodeCall plugin maps its `vm.maxSteps` setting onto the
    // Enclave's `maxToolCalls` cap; set `maxIterations` directly when using
    // Enclave outside the plugin.
    const enclave = new Enclave({ maxIterations: 1000 });
    const result = await enclave.run(`
      let i = 0;
      for (const x of Array(100000)) { i++; }
      return i;
    `);

    expect(result.success).toBe(false);
    expect(result.error?.message).toContain('iteration limit');

    enclave.dispose();
  });
});

CI/CD Integration

# .github/workflows/test.yml
name: Test CodeCall Scripts

on: [push, pull_request]

jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v6
      - uses: actions/setup-node@v6
        with:
          node-version: '20'

      - run: npm ci
      - run: npm test

      # Run CodeCall-specific security tests
      - run: npm run test:security

AgentScript Guide

Complete scripting API, error handling, and best practices

Configuration

Configure meta-tool behavior and limits

Security Model

How scripts are validated and sandboxed

Examples & Recipes

Real-world patterns using the meta-tools

​codecall:search

​Request Schema

​Response Schema

​Examples

​Response Example

​Search Strategy

​codecall:describe

​Request Schema

​Response Schema

​Example

​Security Note

​codecall:execute

​Request Schema

​Response Schema

​Example

AgentScript Guide

​codecall:invoke

​Request Schema

​Response Schema

​Example

​When to Use Invoke vs Execute

​Error Codes Reference

​Script Execution Errors

​Error Handling

​Debugging Guide

​Using console.log

​Interpreting Error Messages

​Development Mode

​Testing Strategy

​Unit Testing AgentScript

​Integration Testing

​Testing Security Rules

​CI/CD Integration

​Related

AgentScript Guide

Configuration

Security Model

Examples & Recipes

codecall:search

Request Schema

Response Schema

Examples

Response Example

Search Strategy

codecall:describe

Request Schema

Response Schema

Example

Security Note

codecall:execute

Request Schema

Response Schema

Example

codecall:invoke

Request Schema

Response Schema

Example

When to Use Invoke vs Execute

Error Codes Reference

Script Execution Errors

Error Handling

Debugging Guide

Using console.log

Interpreting Error Messages

Development Mode

Testing Strategy

Unit Testing AgentScript

Integration Testing

Testing Security Rules

CI/CD Integration

Related