Skip to main content
Glama
Kindora-for-ChatGPT MCP server

Kindora-for-ChatGPT MCP server

by wayanvota

Kindora-for-ChatGPT MCP server

A Python FastMCP server that re-exposes Kindora's public funder and grant tools so you can add them to ChatGPT as a custom MCP connector (Developer mode / Apps). It is a thin proxy: every tool forwards to the live Kindora MCP endpoint and returns the result unchanged.

Easiest install: deploy, then paste the URL into ChatGPT

ChatGPT can't run a local repo; it connects to a hosted HTTPS URL. This server needs no API keys (Kindora's free tier is open), so hosting it is one click, no Terminal, no .env.

Deploy to Render

  1. Click the button, sign in to Render with GitHub, and approve. Render reads render.yaml and builds the container. First deploy takes ~2 minutes.

  2. Copy the service URL Render gives you and add /mcp to the end, e.g. https://kindora-chatgpt-mcp.onrender.com/mcp.

  3. In ChatGPT: Settings -> Connectors (Apps) -> Advanced -> Developer mode, then Create connector, paste the URL, choose No authentication, and save. Enable it from the "+" menu in any chat.

That's the whole install. (Render's free tier sleeps when idle, so the first request after a quiet spell takes 30-60 seconds to wake. Fly.io, below, stays warmer.)

Related MCP server: Personal Assistant MCP Server

What the Kindora MCP does

Kindora is a grant-discovery platform for nonprofits. Its MCP server is a read-only, free-tier service over public IRS 990 / 990-PF filings and Grants.gov opportunities, covering 174K+ US foundations, 32K+ European funders, and 43K+ open grants. The upstream server exposes ten tools across discovery, profile, financials, and reference categories. This wrapper mirrors the nine that matter for ChatGPT (it drops the upstream list_tools helper, which is redundant since MCP clients enumerate tools natively).

Tool

Purpose

search_funders

Find grantmakers by name, cause area, or location

search_open_grants

Find open opportunities / RFPs by topic (foundation + government)

search_funder_jobs

Find open philanthropy jobs at foundations

get_funder_profile

Detailed profile for one foundation (by EIN)

get_990_summary

IRS 990 financial summary and trends (by EIN)

get_foundation_grants

Individual grants a foundation has made (by EIN)

get_funder_stats

Aggregate giving statistics (by EIN)

get_ntee_codes

Browse/search NTEE cause-area codes

health_check

Upstream health probe

All tools are annotated readOnlyHint: true, so ChatGPT runs them without write-confirmation prompts.

Why wrap it instead of pointing ChatGPT at Kindora directly?

ChatGPT Developer mode can connect to any remote MCP server, so a wrapper is optional. It earns its place when you want one endpoint you control: curated tool descriptions, read-only annotations, a place to add auth / rate limiting / logging, header injection for a Kindora key, or to pin a specific upstream URL. If you want none of that, you can skip this and add https://kindora-mcp.azurewebsites.net/mcp/ to ChatGPT directly.

Repository contents

.
├── server.py            # the MCP server (all nine tools)
├── requirements.txt     # runtime dependency (fastmcp)
├── pyproject.toml        # packaging + pytest config
├── test_server.py        # offline tests (no network needed)
├── tests/                # stdio protocol fixture and 20-case E2E suite
├── Dockerfile            # container image for hosting
├── fly.toml              # one-command deploy to Fly.io
├── render.yaml           # one-click deploy to Render
├── .env.example          # configuration reference
├── .github/workflows/    # CI (runs the tests)
├── LICENSE               # MIT
└── README.md

Requirements

  • Python 3.10+

  • A way to host a public HTTPS URL if you want to use it with ChatGPT (Docker + any host, or the included Fly/Render configs).

Get the code

git clone https://github.com/wayanvota/kindora-chatgpt-mcp.git
cd kindora-chatgpt-mcp
pip install -r requirements.txt

Run locally (stdio)

python server.py

Use this for local testing or to add the server to a stdio MCP client (e.g. Claude Desktop).

Run the tests

pip install -e ".[dev]"
python -m compileall -q server.py test_server.py tests
python -m pip check
pytest -q

The suite is offline and deterministic. The unit tests check tool registration, read-only annotations, and upstream-response normalization. The 20-case E2E suite starts the real server as a subprocess, completes MCP initialization over stdio, discovers tools, and calls them through the public protocol. A fixture upstream replaces only the remote Kindora service, so CI needs no credentials or network access.

Run only the protocol-level E2E coverage:

pytest -q tests/test_e2e.py

The categories are labeled U01 through U10 for user workflows and A01 through A10 for validation and abuse cases. They cover discovery, defaults, EIN drilldowns, Unicode, repeated calls, unknown tools, invalid argument types and bounds, inert hostile text, upstream error redaction, and secret safety.

Debug a failing E2E test

Run one category with full output, for example:

pytest -vv -s tests/test_e2e.py::test_a09_upstream_exception_is_generic_and_secret_safe

The fixture process is tests/fixture_server.py. Extend its deterministic responses when adding a tool, then add both a successful protocol test and the most relevant validation or failure case. Do not put a live API key in the fixture, test output, or GitHub Actions. CI stores the JUnit result as a workflow artifact for each supported Python version.

Run as a public HTTP endpoint (required for ChatGPT)

ChatGPT connects to a publicly reachable HTTPS URL, so the server must run with the streamable-HTTP transport behind a public host.

TRANSPORT=http HOST=0.0.0.0 PORT=8000 python server.py
# serves MCP at http://<host>:8000/mcp

Or with Docker:

docker build -t kindora-chatgpt-mcp .
docker run -p 8000:8000 kindora-chatgpt-mcp

Put it behind TLS (a reverse proxy, or a platform like Fly.io / Render / Railway / Cloud Run) so the final URL is https://your-domain/mcp.

One-click hosting

This repo ships configs for two free-tier hosts that give you a public HTTPS URL:

Fly.io (fly.toml):

fly launch --no-deploy --copy-config --name <your-app>
fly deploy
# URL: https://<your-app>.fly.dev/mcp

Render (render.yaml): push the repo to GitHub, then in Render choose New + -> Blueprint and pick the repo. URL: https://<service>.onrender.com/mcp.

Connect it to ChatGPT

  1. Enable Developer mode: Settings -> Connectors (Apps) -> Advanced -> Developer mode. (Available on Plus, Pro, Business, Enterprise, and Edu plans.)

  2. Connectors -> Create / Add custom connector.

  3. Enter your public URL, e.g. https://your-domain/mcp.

  4. Authentication: No authentication (this proxy and Kindora's free tier are open). Add OAuth here only if you put your own auth in front.

  5. Save, then enable the connector in a chat (the "+" / tools menu in the composer).

Developer mode does not require the search/fetch compatibility tools that ChatGPT's Deep Research connectors need; it accepts arbitrary named tools, which is why all nine Kindora tools are exposed directly.

Configuration

All via environment variables (see .env.example):

Variable

Default

Notes

TRANSPORT

stdio

http for ChatGPT

HOST

0.0.0.0

HTTP bind host

PORT

8000

HTTP port

MCP_PATH

/mcp

HTTP path

KINDORA_MCP_URL

https://kindora-mcp.azurewebsites.net/mcp/

Upstream to proxy

KINDORA_API_KEY

(unset)

Forwarded as Authorization: Bearer ... if set

KINDORA_TIMEOUT

60

Upstream call timeout (seconds)

Confirm the upstream URL against your Kindora connector listing. If Kindora moves the endpoint, set KINDORA_MCP_URL and nothing else changes.

Notes and limits

  • Read-only / public data. Nothing here writes. Data is public IRS 990 and Grants.gov.

  • Rate limits. Kindora's free tier is roughly 100 requests/hour for anonymous users; the proxy inherits that.

  • Country fields in grant data reflect the recipient's HQ country, not where program work is implemented.

  • Verify before publishing facts. Treat tool output as leads, not citations: confirm a foundation's current giving and open RFPs against the funder's own filings or site.

Available Tools

9 tools
get_990_summaryA
Read-only

Get an IRS 990 / 990-PF financial summary and year-over-year trends for a foundation.

Returns per-year revenue, end-of-year assets, grants paid, mission text, and computed trends for assets, grants, and revenue.

Inputs: ein (required), years.

get_foundation_grantsA
Read-only

List individual grants a foundation has made, from its 990-PF filings.

Ordered largest-first with aggregate stats. Each grant reports recipient HQ country and US state when known. Note: recipient_country is where the grantee is registered, not necessarily where program work happens.

Inputs: ein (required), year, ntee_code, recipient_country, +3 more.

get_funder_profileA
Read-only

Get a detailed profile for one foundation by EIN.

Returns legal name, location, financials (total assets, annual grants), leadership, classification (foundation type, NTEE), and website. Use search_funders first to find the EIN.

Inputs: ein (required).

get_funder_statsA
Read-only

Get aggregate giving statistics for a foundation.

Lifetime totals, average/median/min/max grant size, top NTEE focus areas, US-state and recipient-HQ-country distribution (each with avg grant size), a non_us_recipient_pct headline, and a year-by-year breakdown. Useful for prospect targeting.

Inputs: ein (required).

get_ntee_codesA
Read-only

Browse or search NTEE classification codes for cause areas.

Call with no arguments to list all 26 major categories. Filter by category letter or search by description to find codes for use with search_funders / get_foundation_grants.

Inputs: category, query.

health_checkA
Read-only

Check that the upstream Kindora service is reachable and healthy.

Inputs: no named parameters.

search_funder_jobsA
Read-only

Find OPEN philanthropy jobs at grantmaking foundations.

Roles involved in giving money away, running philanthropic programs, or executive leadership of philanthropic work. Backed by a weekly scrape of ~50K funder career pages plus LLM classification. Use this for jobs/careers/hiring questions, NOT for grant opportunities (use search_open_grants for those).

For a cause area not covered by the categories below, first call search_funders to get matching funder EINs, then pass them here via funder_eins.

Inputs: query, category, state, country, +8 more.

search_fundersA
Read-only

Find grantmaking organizations by name, cause area, or location.

Searches 174K+ US foundations plus 32K+ European funders from IRS 990 data. Use this to find aligned funders (including ones that fund by relationship, LOI, or annual cycle rather than a live RFP), or when the user names a specific funder. For active/open opportunities use search_open_grants instead.

Inputs: query, state, city, ntee_code, +8 more.

search_open_grantsA
Read-only

Find OPEN grant opportunities and RFPs by topic or cause area.

The primary tool for finding grants. Searches 172K+ foundation programs and federal Grants.gov opportunities across program names, descriptions, focus areas, and beneficiary types, with stemming and natural-language matching.

Query syntax: natural language ("affordable housing for seniors"), quoted phrases ('"after school"'), and exclusion ("education -higher"). Omit query to browse broadly open programs sorted by upcoming deadline.

Inputs: query, focus_area, agency, state, +7 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.4/5.0

    Scored across 9 tools

    Disambiguation5/5

    Each tool serves a clear, distinct purpose: foundation financials, grants, profiles, stats, search for funders/grants/jobs, NTEE codes, and health check. Descriptions explicitly differentiate between similar tools like search_funders, search_open_grants, and search_funder_jobs.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun snake_case pattern (e.g., get_990_summary, search_funders, health_check). No mixing of conventions like camelCase or inconsistent verb forms.

    Tool Count5/5

    9 tools is well-scoped for a philanthropy-focused server, covering all major user needs: exploring foundations, grants, jobs, and NTEE codes, with a health check. Not too many to overwhelm, and each tool earns its place.

    Completeness5/5

    The tool set covers the full lifecycle of philanthropy research: finding funders and NTEE codes, retrieving detailed profiles and financial trends, listing individual grants, searching open grants and jobs, and system health. No obvious gaps for the intended use.

    Maintenance

    ActivityMaintained
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      D
      maintenance
      An educational MCP server example built with FastMCP that demonstrates how to expose tools, resources, and prompts to AI clients. Provides a learning foundation for building MCP servers with Python and integrating them with AI applications like IDEs and chatbots.
      -
    • A
      license
      Not graded
      quality
      D
      maintenance
      A FastMCP-based MCP server for the R2R API, enabling integration with document management, knowledge graphs, and RAG systems through automatically generated tools and resources.
      MIT