Fieldproxy MCPサーバー

Fieldproxy MCP server

The official Model Context Protocol server for Fieldproxy field service management: connect Claude, ChatGPT, Copilot, Cursor or any MCP client to jobs, dispatch, invoices and quotes, with a person confirming every change.

The server is hosted by Fieldproxy (streamable HTTP). This repository holds the docs, client configs, runnable examples and a small stdio bridge for clients that cannot speak HTTP.

• Overview: https://www.fieldproxy.ai/mcp
• Developer docs: https://www.fieldproxy.ai/developers
• Sandbox page: https://www.fieldproxy.ai/developers/sandbox

Try it in 60 seconds

There is a public, read-only sandbox you can connect to right now. It is a sample HVAC company, Summit Heating & Air, with made-up data: about 2,076 jobs, 435 customers in New York, 12 technicians, invoices, estimates, equipment and service agreements.

Address	https://api-uk.fieldproxy.ai/mcp
Header	Authorization: Bearer fpmcp.eleqia.pZWm5OR4qsYqP1Wb0DKsxm9bfs_vdI-DCHL5JJMPlxc
Scope	data:read only. No writes, no email, no SMS.
Key expires	4 February 2027 (rotated before then; this README and the sandbox page show the current key)

The key is public on purpose. It can only read the sample data.

Claude Code

claude mcp add --transport http fieldproxy-sandbox https://api-uk.fieldproxy.ai/mcp \
  --header "Authorization: Bearer fpmcp.eleqia.pZWm5OR4qsYqP1Wb0DKsxm9bfs_vdI-DCHL5JJMPlxc"

Cursor (.cursor/mcp.json), or a project .mcp.json

{
  "mcpServers": {
    "fieldproxy-sandbox": {
      "url": "https://api-uk.fieldproxy.ai/mcp",
      "headers": { "Authorization": "Bearer fpmcp.eleqia.pZWm5OR4qsYqP1Wb0DKsxm9bfs_vdI-DCHL5JJMPlxc" }
    }
  }
}

Any MCP client that speaks streamable HTTP

{
  "type": "http",
  "url": "https://api-uk.fieldproxy.ai/mcp",
  "headers": { "Authorization": "Bearer fpmcp.eleqia.pZWm5OR4qsYqP1Wb0DKsxm9bfs_vdI-DCHL5JJMPlxc" }
}

Clients that only run local (stdio) servers, such as Claude Desktop

Use mcp-remote as a bridge:

npx -y mcp-remote https://api-uk.fieldproxy.ai/mcp \
  --header "Authorization: Bearer fpmcp.eleqia.pZWm5OR4qsYqP1Wb0DKsxm9bfs_vdI-DCHL5JJMPlxc"

In claude_desktop_config.json see examples/claude_desktop_config.json (the key goes in env so the space in the header survives on Windows).

Or use this repository's own bridge, published as @fieldproxy/mcp. With no settings it connects to the sandbox:

{
  "mcpServers": {
    "fieldproxy": { "command": "npx", "args": ["-y", "@fieldproxy/mcp"] }
  }
}

No client at all

Run one of the examples. They call initialize, then run_sql, and print the rows. No packages needed.

node examples/run_sql.mjs          # Node.js 18+
python3 examples/run_sql.py        # Python 3.8+
python3 examples/run_sql.py "SELECT name, role FROM st_technicians WHERE active"

Or with curl:

curl -s https://api-uk.fieldproxy.ai/mcp \
  -H "Authorization: Bearer fpmcp.eleqia.pZWm5OR4qsYqP1Wb0DKsxm9bfs_vdI-DCHL5JJMPlxc" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"run_sql","arguments":{"sql":"SELECT status, count(*) FROM st_jobs GROUP BY status"}}}'

Things to ask

• "Call describe_schema, then tell me how many jobs are in each status."
• "Which job types do we run most often?"
• "List the scheduled jobs in Glen Cove with the customer's phone number."
• "Which technicians ran the most visits in the last 90 days?"
• "Which open estimates have had no follow-up for 30 days or more?"
• "Which customers own equipment more than 12 years old? Group it by customer."
• "How long is the drive from Garden City to Glen Cove?"
• "Plan the best route through tomorrow's scheduled visits."
• "Export every pending invoice to CSV."

The sandbox key is limited to the sample tables, so run_sql refuses SQL functions that could read other tables, such as sum() and avg(). count(), max(), GROUP BY, joins and date filters work. A key on your own workspace does not need that limit.

Sandbox REST API

The same company is also open on the REST API, read only:

curl -X POST https://api-uk.fieldproxy.ai/api/read \
  -H "x-api-key: eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJ1c2VyaWQiOiJ0ZXN0IiwiZW1haWwiOiJ0ZXN0QGV4YW1wbGUuY29tIiwidGVuYW50X2lkIjoiZWxlcWlhIiwidHlwZSI6ImFwaV9rZXkiLCJqdGkiOiI0YzliYjgwZi0yY2U2LTQ4ODYtYjk2NS1kMjA2YWU4ODczMWEiLCJpYXQiOjE3OTM5NTU3NTEsImV4cCI6MTgyNTQ5MTc1MX0.SgIxGkXEqMPbikAoX4D7H4coOUvnlrsJYb-uA8_ixNA" \
  -H "content-type: application/json" \
  -d '{"tableName": "st_customers", "condition": {"id": 1}}'

Sandbox tools

The sandbox key exposes 21 tools (read live from tools/list on 6 November 2026). The tool list is built per key, so the tools you see depend on the key's scopes and limits.

Tool	What it does
describe_schema	Lists the tables the key may read, or the columns of named tables.
run_sql	Runs one read-only SELECT (up to 500 rows, 5-second limit).
search_freeform	Runs a read-only SELECT or WITH query when no other tool fits.
list_distinct_values	Distinct values in a column with row counts, such as job statuses.
explain_change	Explains why a number changed between two periods, split by a column.
export_csv	Saves a SELECT result as a CSV file (up to 10,000 rows) and returns a link.
open_record	Returns a link that opens one record in the Fieldproxy web app.
list_record_files	Lists the photos, PDFs and signatures stored on one record.
generate_pdf	Turns HTML (a quote, job report, summary) into a PDF link.
list_dashboards	Command Center dashboards the key can see.
view_dashboard	Each tile of a dashboard with its current data.
list_alerts	Data alerts and when they last ran.
list_scheduled_reports	Scheduled reports, cadence, next and last run.
workspace_usage	Record and seat limits and this month's AI credits.
geocode_address	Coordinates for up to 10 addresses.
reverse_geocode	The street address nearest to a coordinate.
drive_time	Road drive time and distance between two places.
calculate_route	The best order to visit up to 50 stops.
get_widget_docs	Fieldproxy's app-building reference.
get_automation_docs	Fieldproxy's automation reference.
log_turn	Saves the conversation turn to the user's Command Center history.

The server also offers seven ready-made prompts: daily_ops_briefing, plan_the_schedule, triage_my_inbox, why_did_this_automation_fail, customer_360, build_an_app, build_a_dashboard.

Connect your own workspace

A full workspace key has 170+ built-in tools, plus any SQL tools the workspace saves for its AI.

Area	What it can do
Read	Schema and SELECT-only SQL. Customers, jobs, technicians, invoices, quotes, contracts. AR ageing, technician performance, funnels.
Write	Propose a change, confirm it, undo it. CSV import and export, record history, revert a record, restore deleted records.
Dispatch and maps	Spread jobs across technicians, plan daily routes, apply the schedule. Drive times, nearest technicians, live locations, job sign-off links.
Comms	Search the inbox and read threads. Draft emails and SMS from templates, then send after you confirm. Comments, notifications, follow-ups and calls.
Sales documents	Preview and issue proposals, generate estimates and PDF documents.
Files and photos	Read PDF, Word, Excel, CSV and email files. View job photos.
Monitoring	Dashboards, alerts, scheduled reports, the activity feed, usage.
Automation operations	Run a rule, retry or dismiss failed runs, decide approvals.
AI oversight (admins)	Every AI connection in the workspace, what each one did, usage by key, person and tool, revoking access.
Build	Apps, tables and columns, automations, voice and text agents, document and email templates. Saved as drafts; new automations and agents are saved switched off.
Sandbox copy	Copy the workspace, build in the copy, then plan and confirm a promotion to live.

Addresses

Each workspace lives in one region. Your address is shown in Settings, AI assistant.

Region	Address
US East	https://api-us-east-1.fieldproxy.ai/mcp
Canada	https://api-canada.fieldproxy.ai/mcp
UK	https://api-uk.fieldproxy.ai/mcp
India	https://api-india-1.fieldproxy.ai/mcp
Norway	https://api-norway.fieldproxy.ai/mcp
New Zealand	https://api-nz.fieldproxy.ai/mcp

AI assistant access is off by default. Fieldproxy switches it on per workspace. Ask your Fieldproxy contact, or book a demo.

OAuth or a key

• OAuth (sign in). Claude, ChatGPT and Microsoft Copilot Studio connect with OAuth and dynamic client registration: add the address, sign in with workspace ID, email and password, and tick the scopes. The sign-in lasts 1 hour and renews itself for 30 days. Guides: Claude, ChatGPT, Microsoft 365 Copilot and Teams.
• Key. Make one in the web app under Settings, AI assistant, New key. Keys start with fpmcp. and are shown once. Send it as Authorization: Bearer fpmcp.…. An administrator can set up a key for someone else; that person claims it once from their own screen, so the administrator never sees it.

Client setup with a key

Claude Code:

claude mcp add --transport http fieldproxy https://api-us-east-1.fieldproxy.ai/mcp \
  --header "Authorization: Bearer fpmcp.YOUR_KEY"

OpenAI Codex:

export FIELDPROXY_MCP_KEY="fpmcp.YOUR_KEY"
codex mcp add fieldproxy --url https://api-us-east-1.fieldproxy.ai/mcp --bearer-token-env-var FIELDPROXY_MCP_KEY

To sign in instead of using a key, add the server without --bearer-token-env-var, then run codex mcp login fieldproxy.

Stdio bridge:

FIELDPROXY_MCP_URL=https://api-us-east-1.fieldproxy.ai/mcp \
FIELDPROXY_MCP_KEY=fpmcp.YOUR_KEY \
npx -y @fieldproxy/mcp

The bridge sends a key; for OAuth over stdio use npx -y mcp-remote <address> without a header and it opens the sign-in page.

Config files for each client are in examples/:

File	Client
claude-code.sh	Claude Code
claude_desktop_config.json	Claude Desktop (stdio via mcp-remote)
cursor-mcp.json	Cursor, .mcp.json
vscode-mcp.json	GitHub Copilot in VS Code (.vscode/mcp.json)
codex-config.toml	OpenAI Codex (~/.codex/config.toml)
run_sql.mjs	Node.js script
run_sql.py	Python script

Scopes

Scope	Requires	Unlocks
data:read	AI assistant access on for the workspace, and the user has Command Center access	Schema, ready-made reports, read-only SQL
data:write	Tables the workspace lets the AI write, and no approval gate on them	Changes to records, confirmed by a person and undoable; email and SMS drafts, sent on confirm
build	An administrator, plus Allow build switched on for the workspace	Fields, tables, screens, automations, voice and text agents, sandbox copies

The user's own Command Center access sets the ceiling. A key can be limited further.

Security model

• Confirm before change. Every write, send or action is a proposal a person confirms within 30 minutes. Writes can be undone. More than 5 rows needs bulk permission, and more than 1,000 is refused.
• Scoped keys. Limit a key to certain tables, certain rows ("only the North region's jobs"), hidden columns, insert/update/delete, and an expiry from 1 hour to 90 days. Postgres enforces the row limits, and hidden columns are removed from every result.
• A tool list built per person. The tools the AI sees are assembled for each request from the key's scopes, the person's role, the workspace settings and the key's limits. A tool missing from the list is refused even when called by name.
• Revocation on the next call. Nothing is kept between requests. Each one is authorised from scratch, so a revoked or expired key is refused at once.
• Audit. Every call is logged with the tool, key, person and outcome. Administrators see every key and connected app and can revoke any of them.
• Sends are limited. Email goes only from the user's own mailbox, at most 20 sends an hour per key, inside the workspace's quiet hours. A sandbox copy never sends.
• Builds are drafts. Publishing is blocked if the linter, a SQL check or a navigation check finds a new error. Every version is kept for rollback.

The stdio bridge (@fieldproxy/mcp)

bin/fieldproxy-mcp.js is a small bridge with no dependencies (Node.js 18+). It reads MCP messages on stdin, posts each one to the server, and writes the replies to stdout.

npx -y @fieldproxy/mcp --help

  --url <url>       MCP address (env FIELDPROXY_MCP_URL)
  --key <key>       fpmcp.… key (env FIELDPROXY_MCP_KEY)
  --header "N: v"   extra HTTP header, repeatable
  --sandbox         force the public sandbox

npm test starts the bridge and runs initialize, tools/list and one run_sql against the sandbox.

Links

• Overview: https://www.fieldproxy.ai/mcp
• Developer docs: https://www.fieldproxy.ai/developers
• Sandbox: https://www.fieldproxy.ai/developers/sandbox
• Which field service software has an MCP server: https://www.fieldproxy.ai/mcp#which-fsm-has-mcp
• MCP Registry entry: io.github.Fieldproxy/fieldproxy-mcp
• Book a demo: https://www.fieldproxy.ai/book-demo?source=github-mcp